{"metadata":{"bundle_type":"episode_pack","bundle_version":"prompt24_v1","workspace_slug":"orbital","episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","exported_at":"2026-07-01T18:11:11.911255Z"},"summary":{"content_asset_count":21,"transcript_segment_count":200,"asset_types":{"content_calendar_item":3,"newsletter_summary":1,"social_post":3,"hook":3,"quote_card":4,"clip_candidate":4,"episode_theme":3},"ranked_theme_ids":[],"theme_snapshot_ids":[]},"episode":{"id":"fe4a2424-e193-43a6-b472-dadb11b69431","source_id":"18dc6d14-f8ca-47a4-9e12-a42b7b0284ba","source_slug":"yt-nw8yN8Nhgfc-d9654309","transcript_document_id":"1215046d-0e0a-4358-b348-02488f3cb423","raw_asset_id":"838b9558-711d-4be4-974e-1fcdde39dddf","title":"Get started with generative AI and agents in Azure | AI-901 | Episode 5","description":"Episode 5 of 14 For the full video series, click here: https://aka.ms/AI-901onYouTube Choosing the right AI model doesn’t have to be guesswork. In this video, you’ll discover how Azure AI Foundry helps you choose the right AI model, test it in the Playground without writing code, and transform it into an intelligent agent capable of taking real actions. From model selection to building intelligent agents, this is your practical guide to creating smarter, safer AI solutions. Immerse yourself in our rich, interactive materials at your own pace with self-directed learning: https://aka.ms/AI-901onLearn 00:00 Video Start 02:01 Microsoft Foundry Models 06:30 Using a generative AI model 10:58 Creating agents 15:26 Demo: Get started with Microsoft Foundry","external_url":"https://www.youtube.com/watch?v=nw8yN8Nhgfc","status":"published","published_at":"2026-06-29T15:00:30Z","transcript_segment_count":200,"content_asset_count":21,"details_json":{"file_name":null,"published_at":"2026-06-29T15:00:30+00:00","transcript_format":"youtube_captions"},"latest_transcript_segments":[{"id":"04b2aa5e-b2f5-4f10-ac90-af963f3fc8ac","segment_index":0,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"[ MUSIC ]"},{"id":"da5799ee-b998-4a21-b689-8e10b83aa950","segment_index":1,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"SARAH ALLALI: Okay."},{"id":"9bdf9a5a-c5c8-46d8-bb90-ef39ebdb3281","segment_index":2,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"I want to build an AI assistant, one that can answer questions,"},{"id":"67fe5a81-0eb9-4ac1-900a-e0e194c4e135","segment_index":3,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"check information, and even take actions."},{"id":"0812b5c7-c058-49e1-b3aa-aaaa83ebfbe3","segment_index":4,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"But now I have a bigger question."},{"id":"c7e61c41-ca4a-4520-acdf-059306006904","segment_index":5,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Which model should I choose?"},{"id":"40455371-1f76-4ede-9d7f-a1a516fefda8","segment_index":6,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"There are so many options, different models,"},{"id":"7d1480e9-4609-4fbf-b831-080c64bfdeca","segment_index":7,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"different capabilities, different providers;"}],"content_asset_counts":{"content_calendar_item":3,"newsletter_summary":1,"social_post":3,"hook":3,"quote_card":4,"clip_candidate":4,"episode_theme":3}},"source":{"id":"18dc6d14-f8ca-47a4-9e12-a42b7b0284ba","workspace_id":"d9654309-c206-4820-9522-1886720e58c4","name":"Get started with generative AI and agents in Azure | AI-901 | Episode 5","slug":"yt-nw8yN8Nhgfc-d9654309","source_type":"youtube_video","enabled":true,"base_url":"https://www.youtube.com/watch?v=nw8yN8Nhgfc","feed_url":null,"domain":"youtube.com","jurisdiction":null,"authority_tier":"tier_d","source_class":"exploratory","access_posture":"fully_fetchable","discovery_confidence":null,"original_discovery_confidence":null,"approval_basis":null,"promotion_path":null,"discovery_context_json":{},"content_type_label":null,"cadence":null,"fetch_method":null,"last_ingested_at":"2026-06-30T06:10:54.443637Z","archived_at":null,"last_run_status":"succeeded","document_count":1,"recent_document_count":1,"policy":{"layer":null,"fetch_strategy":null,"effective_cadence":"weekly","is_event_driven":false,"last_checked_at":"2026-06-30T06:10:54.443637Z","last_changed_at":"2026-06-30T06:10:54.443637Z","next_check_at":"2026-07-07T06:10:54.443637Z","change_state":null,"cache_etag":null,"cache_last_modified":null,"cache_status":null,"stale":false,"due_now":false,"due_reason":"within_cadence"},"evidence_posture":{"origin_lane":"search_council_exploratory","source_class":"exploratory","trust_posture":"exploratory","evidence_class":"source_evidence","access_posture":"fully_fetchable","promotion_status":"candidate","admissibility_status":"context_only","evidence_floor_status":"supporting_floor","evidence_floor_reason":"Exploratory and promoted material remains below the primary evidence floor until it earns stronger trust.","summary":"Exploratory and promoted material stays below the primary evidence floor.","reasons":["Explicit access posture hint: fully fetchable.","This source is still exploratory and has not cleared the promotion floor.","Exploratory sources remain context-only until they are promoted through discovery.","Exploratory and promoted material remains below the primary evidence floor until it earns stronger trust."]},"source_reliability":{"score":42.6,"band":"guarded","summary":"Source reliability is guarded at 42.6/100.","reasons":["Authority tier is tier_d, contributing to a guarded reliability posture.","Access/admissibility posture is context_only, so Orbital scores reliability with that trust ceiling in mind.","The source has 1 documents in Orbital, including 1 in the recent window.","This source came through discovery/promotion, so reliability is intentionally capped below a strong curated source unless history accumulates."],"factors":[{"name":"authority_tier","value":7.0,"reason":"Higher authority tiers carry more baseline reliability."},{"name":"source_class","value":4.0,"reason":"Curated and manually approved sources start from a stronger trust base than exploratory promotions."},{"name":"admissibility","value":4.0,"reason":"Access and admissibility posture should raise or limit downstream reliance."},{"name":"coverage_history","value":2.6,"reason":"Sources with more durable document history are more reliable than one-off appearances."},{"name":"operational_health","value":6.0,"reason":"Recent ingestion success is a bounded proxy for source stability."},{"name":"overclaim_risk","value":0.0,"reason":"Low-access or lightly observed sources should be scored more cautiously."}]},"lifecycle":{},"config_json":{"title":"Get started with generative AI and agents in Azure | AI-901 | Episode 5","video_id":"nw8yN8Nhgfc","channel_id":"UCFs5990vNpA2C2ijzgiJ59Q","fetched_at":"2026-07-01T06:08:11.917460+00:00","description":"Episode 5 of 14\nFor the full video series, click here: https://aka.ms/AI-901onYouTube\n\nChoosing the right AI model doesn’t have to be guesswork. In this video, you’ll discover how Azure AI Foundry helps you choose the right AI model, test it in the Playground without writing code, and transform it into an intelligent agent capable of taking real actions. From model selection to building intelligent agents, this is your practical guide to creating smarter, safer AI solutions. \n\nImmerse yourself in our rich, interactive materials at your own pace with self-directed learning: https://aka.ms/AI-901onLearn\n\n00:00 Video Start\n02:01 Microsoft Foundry Models\n06:30 Using a generative AI model\n10:58 Creating agents\n15:26 Demo: Get started with Microsoft Foundry","topic_seeds":["AI Systems Discovering Non-Obvious Model Configurations"],"caption_kind":"manual","channel_name":"Microsoft Learn","published_at":"2026-06-29T15:00:30Z","source_universe":{"cadence":"weekly","warnings":[],"feed_urls":[],"source_type":"youtube_video","sitemap_urls":[],"allowed_paths":[],"blocked_paths":[],"crawl_posture":"manual_approval","domain_family":"youtube.com","freshness_sla":"weekly","source_family":"video_media","trust_posture":"community","authority_tier":"tier_d","branch_relevance":[],"historical_yield":{},"rejection_history":[],"source_family_raw":null,"audience_relevance":[],"contamination_history":[],"robots_unavailable_policy":"fail_closed"},"transcript_text":"[ MUSIC ]\nSARAH ALLALI: Okay.\nI want to build an AI assistant,\none that can answer questions,\ncheck information, and\neven take actions.\nBut now I have a\nbigger question.\nWhich model should I choose?\nThere are so many options,\ndifferent models,\ndifferent capabilities,\ndifferent providers;\nand the choice actually matters\nbecause the model I choose will\ndirectly impact how my agent\ngoing to behave.\nSo how do I choose\nthe right model\nfor the right scenario\ninstead of just guessing?\nWhat if I had a single place\nwhere I can explore, compare,\ntest models before using them?\nThat's exactly what\nMicrosoft Foundry comes in.\nIf you are wondering how to\nchoose models and use them\nto build a diligent agent,\nyou are in the right place.\nHi. I'm Sarah Allali, a\nsenior technical trainer\nat Microsoft specializing in AI\nCopilot and building intelligent\nand secure application\non Microsoft Azure.\nI have a PhD in computer science\nand 12 years teaching\ntraining experience.\nThis course covers 12\nsessions where we'll dive\ninto CoreAI concepts and Azure\nand learn how to turn ideas\ninto real applications\nusing modern tools\nlike Microsoft Foundry.\nIn the first session, we\nwill see how to get started\nwith generative AI\nand agents in Azure.\nI will show you how Microsoft\nFoundry helps you discover,\nevaluate, and use AI models\nto build real\napplications and agents.\nUp to now, we have talked\nabout models, agents,\nand how they work together.\nNow let's answer a very\npractical question.\nWhere do these models come from,\nand how do we choose\nthe right one?\nThat's exactly where\nMicrosoft Foundry comes in.\nThink of Microsoft Foundry\nas a created marketplace\nand delivery platform\nfor AI models.\nInstead of randomly picking\na model and hoping it works,\nFoundry gives you a catalog\nwhere models are clearly\ndescribed, categorized,\ngoverned, and delivered in a\nsafe, enterprise-ready way.\nYou are not guessing.\nYou are choosing intentionally.\nAt Microsoft Foundry, we provide\ntwo big model categories,\nmodels sold directly by\nMicrosoft and models\nfrom partner and community.\nModels sold directly by\nMicrosoft are models\nthat are hosted by\nMicrosoft and built\nthrough your Azure subscription.\nThey come with enterprise\nsupport and SLAs,\nfollow Microsoft responsible\nAI standards that we talked\nabout in a previous session.\nThese include all Azure OpenAI\nmodels and selected models\nfrom trusted providers that\nMicrosoft manages directly.\nFor example, if you are a\ncompany building an internal HR\nassistant and you want\nenterprise security,\npredictable billing,\nand Microsoft support,\nyou will typically start with\nAzure OpenAI models in Foundry.\nThe partner community models\ngives access to models\nfrom Research Lab, open\nsource communities,\nand specialized AI providers.\nThese models are provided by\ntrusted third-party partners,\npurchased under the partners'\nterms, and are useful for niche\nor specialized use cases.\nFor example, you are building an\nAI system for medical imaging\nor legal document analysis.\nA partner or community\nmodel may be a better fit\nthan a general purpose\nlanguage model\nfor some specific scenarios.\nFoundry doesn't\njust list models.\nIt helps you compare\nthem intentionally based\non real decision factors.\nAnd let's break those down.\nFirst, deployment type meta,\nwhere the work happens.\nThis answer questions like\nwhere is your data processed?\nHow is the model hosted?\nHow do you pay for it?\nFor example, a public\ncustomer-facing chatbot may use\na fully managed\ncloud deployment.\nA sensitive internal tool may\nrequire stricter hosting rules.\nThe second factor is\nversioning and updates.\nThese depends on\nyour preference.\nDo you prefer stability, or\ndo you prefer innovation?\nModels evolve, and Foundry lets\nyou see the model version,\nwhether it is auto updated or\nstay fixed for consistency.\nFor example, if you are in\na finance or healthcare,\nyou may prefer a fixed version.\nIf you are experimenting, you\nmay welcome automatic upgrades.\nThe third factor is\nabout the rate limits.\nThis is all about how\nfast and how often.\nEvery model has limits, such as\nthe maximum tokens per minute\nand the throughput constraints.\nSo, for example, a high-volume\ncustomer support bots needs\nhigher rate limits than a\nsmall internal assistant.\nThe last factor is\nabout guardrails\nand responsible AI policies.\nAnd this is critical.\nFoundry models come with\nbuilt-in responsible AI policy,\ncontent safety control,\nand usage guidelines.\nFor example, you don't want your\nAI assistant generating unsafe\nor inappropriate content.\nFoundry helps apply\nguardrails by design,\nnot as an afterthought.\nNow that we understand\nmodels and agents,\nlet's look at how we actually\nuse a generative AI model\nin practice.\nAnd the most important\nrule here is simple:\nBefore you write\ncode, you experiment.\nThat's exactly why Microsoft\nFoundry provides the\nModel Playground.\nThink of the Model\nPlayground as a sandbox.\nIt is a safe place where you\ncan try different models,\ntest prompts, adjust settings,\nand see result instantly,\nall without writing a\nsingle line of code.\nFor example, you want an AI\nto help employees write\nprofessional emails.\nIn the playground, you can type\nor write a polite email asking\nfor a deadline extension.\nYou see the result immediately.\nTweak the wording and refine the\ntone before you build anything.\nWhy the playground\nis so valuable?\nBecause it helps\nyou to test prompt\nby understanding what works\nand what doesn't work.\nTune parameter by controlling\ntone, length, creativity,\nand also capture configuration\nby saving exact\nsetting before coding.\nThis avoids a very common\nmistake, jumping straight\ninto code and debugging\nprompt at the same time.\nWhen you work with a\ngenerative AI model,\nwhether in the\nplayground or in code,\nyou always configure three\nmain things: instructions,\ninput, and parameters.\nInstruction, or system\nprompt, sets the rule\nand the rules for the model.\nThink of it as giving the\nAI a job description.\nFor example, for an HR\nagent, we would tell it,\n\"You are a helpful HR assistant.\nYou respond professionally\nand clearly.\nYou don't give any legal advice.\nThis frames how\nthe model behaves.\nNext, we have input,\nor user prompt.\nThis is what the\nuser actually asks.\nGood prompts are always clear,\nspecific, and context-rich.\nFor example, rather than\nwriting, Write an email,\nin the prompt, you would say,\nWrite a friendly email\nreminding a colleague\nabout a meeting\ntomorrow at 10am.\nThis prompt will get you\na much better result.\nFinally, the parameters.\nThe parameters is all about\nhow the model responds.\nParameters help fine\ntune the response\nby using factors\nsuch as creativity.\nIf you put it low, you\nwill get you a precise,\npredictable answers.\nAnd, if you put it high,\nyou will produce a more\ncreative, varied response.\nAnother factor is the\nresponse length, like a short\nor maybe a long or maybe\nmore detailed explanation.\nSo what is the best\nconfiguration for the factors?\nThis depends on the context.\nFor example, for\na legal summary,\nlow creativity is better;\nand, for marketing copy,\nhigh creativity works well.\nOnce you are happy\nin the playground,\nmoving to code is\nstraightforward.\nFor example, let's look\nat this without diving\ninto technical details.\nIn the slide, you have a\ntypical Python example\nthat includes an endpoint\nwhere the model lives,\nan API key to\nauthenticate securely,\nand the prompt\nsubmission in the input,\nthe same prompt you tested.\nWhat is important to know the\nexact same concept applied\nin Python, JavaScript,\nC#, or any language.\nOnly the syntax is going\nto change, not the logic.\nAnd yes; you use OpenAI or APIs\nthrough Azure Microsoft Foundry\nto consume these models in code.\nUp to now, we have been working\nwith generative AI\nmodels testing prompt,\ntuning parameters,\nand understanding how\nmodels generate responses.\nThe natural next step is\nturning a model into an agent.\nThat's exactly what the\nMicrosoft Foundry Agent Service\nis for.\nWhat does creating an\nagent really mean?\nIn simple terms, a model answers\nquestions; an agent is designed\nto get something done.\nWhen you create an\nagent in Foundry,\nyou take a model\nconfiguration that works;\nsave it as a named agent; give\nit persistent instruction; and,\nfinally, connect it to\ntools and knowledge\nso the agent doesn't forget its\nrole every time you talk to it.\nOne key difference from models\nis that agents remember\nwho they are and what\nthey are supposed to do.\nTheir instructions act\nlike a job description.\nWhat they help with,\nwhat tool to use,\nand what they should not do,\nthis makes agent consistent\nand reliable across\nconversations.\nFor example, you create an\nagent called Project Assistant\nwith instructions like, You\nhelp me track project progress\nand prepare clear status report.\nYou do not make\nbusiness decisions.\nEvery time you use this agent,\nit behaves consistently.\nIt doesn't forget who it is.\nAgents are not limited\nto what's in the prompt.\nUsing the Foundry Agent Service,\nagents can serve for knowledge,\nfor example, project files,\ndocument, internal data sources,\nor approved web sources.\nFor example, instead of you\npasting content into a chat,\nyou can ask, what were\nthe main decisions\nfrom last week's meeting.\nThe agent searches the meeting\nnotes and extracts the answers.\nThis is data search for\nknowledge, not guessing.\nIn addition, agents\ncan take actions.\nThis is where agents\nclearly go beyond models.\nAgents can use action\ntools to automate tasks\nsuch as sending emails, creating\nsummaries, updating recourse,\nor generating structured\noutput or code.\nFor example, a meeting assistant\nagent can read meeting notes\nand then generate a summary and\nfinally send it to participants.\nJust like with models, Foundry\ngives you a safe agent\nplayground where you can test\nyour agent's instructions,\ntry different tools, see how it\nbehaves, and fix issues early,\nall before connecting it to\nreal system or writing code.\nOnce your agents behave the way\nyou want, now you can connect it\nto your application and\nstart consuming it via APIs.\nJust like models,\nyou can connect\nto your agent using the Foundry\nProject API via secure endpoints\nand authentication obviously.\nAgain, from an app perspective,\nagents are consumed\nthrough APIs exactly\nlike models.\nThe difference is what the agent\ncan do once it gets requested.\nTo summarize what I have said\nso far, creating an agent\nwith the Foundry Agent\nService means you start\nfrom a model that works.\nYou add persistent instructions.\nYou connect knowledge and tools.\nYou enable automation.\nAnd, finally, you still consume\neverything through APIs.\nAnd one more time, remember\nmodels answer questions while\nagents help get the work done.\nNow, let's do a quick exercise\nwhere we will continue\nour discovery\nof the Microsoft Foundry portal.\nWe will experiment with\nthe deployed model\nin the playground, and then\nwe will observe the effect\nof system prompts and\nalso the parameter.\nFinally, we will create an\nagent with a file search tool.\nYou can also practice this\nexercise in your own environment\nby accessing the lab steps\nin the link that you can see\non the screen or by visiting\nthe AI-901 Learning Path\nthrough the Microsoft\nLearn platform.\nLet's go.\nIn this exercise, we're going to\nuse Microsoft Foundry to deploy\nand explore generative\nAI models, and.\nWe're going to look into\nthe agent, knowledge tools,\nhow to answer\nquestions, and so on.\nSo here's starting from the\nMicrosoft Foundry portal,\nand we are staying\nthe same project\nthat we created in\nprevious session.\nWe are going to also\nsay use the same model.\nSo, if you haven't done so, you\ncan get back to that session\nand see the demo of how we\ncould possibly create an agent\nfrom the discovery box; and then\nyou will find your model ready\nto be used.\nSo here's the GPT 4.1 mini model\nthat I'm going to continue with\nand use in this project\nthat focus on agent.\nYou can also practice your\nown, in your own environment,\nthe exercise lab that\nyou can find the steps\nin the AI-901 course.\nSo, once we have our project\nready, our model ready, well,\nlet's start out\nworking on our agent.\nNow, here I'm clicking on my\nmodel and which is GPT 4.1 model\nthat we created before.\nAnd I'm going to try\nout with while talking\nto the model, see\nhow it behaves.\nWho was, for example,\nAda Lovelace?\nI get an answer information\nabout who is that person.\nI can see that it used\nthe model GPT 4.1 mini.\nSo I can see also all the\ninformation that I needed,\nfrom the model to the time\nspent to the token used.\nAnd, also, I can copy the result\nor maybe delete all\nthis discussion\nand answer that I get.\nIf I look at the token,\nI have two things.\nYou can see the\ninput, the output.\nThe input depends to your\nprompt, to whatever you're going\nto feed the model; and also the\noutput depends to the answer\nthat you get from the model.\nSo let's keep up\nthe conversation.\nLet's ask it, tell me about\nher work with Charles.\nSo here's it detected that\nI still talking about Ada,\nso give me information about\nAda and Charles and their work.\nAnd I can see also the\nresult here for the token.\nIt's different.\nSo you see the input\nis a little bit bigger\nbecause also we have\nthe discussion history.\nThe output's a\nlittle bit longer,\nso our output is also\na little bit bigger\nwhen it comes to token.\nNow let's start a new chat.\nAnd now again to ask, tell\nme about the ELIZA chatbot.\nSo it's providing me the\ninformation about ELIZA chatbot.\nI see the information here.\nI see the number of token used.\nInput, again, shorter;\nno history discussion.\nOutput a little bit longer.\nWe get a longer, little\nbit longer output in here.\nI can keep up the conversation\nagain, say, how does it compare,\nfor example, with modern LLMs\nand see if it's going to keep\nup the discussion and follow up\nwith or connect the conversation\nwith the ELIZA chatbot that we\ntalked about it previously.\nSo you see here\nELIZA and modern LLM\nand give me a longer\ncomparison here.\nAlso, we have a table,\ncomparison table.\nShe's great.\nAnd can see also for the token\nhow much the response token get.\nYou also have here, you can\nsee here the code section.\nSo, in the code section, so\nif you are a developer going\nto really start coding, well,\nyou need to look at\nthe code section.\nAnd in the code\nsection you can see\nat the top we do have\ndifferent configuration\nthat you could change or work\non to see the code or the piece\nof code that reflect\nthe configuration\nthat you're going\nto put in place.\nTo start, we have the API.\nSo we have the completion\nand the response API.\nThere is two.\nWe have also the language.\nHere we're selecting Python.\nWe have other language.\nYou have the SDK, also\nusing the OpenAI SDK.\nAnd also these\nthree dots help us\nto do more configuration related\nto, for example, authentication.\nAnd whenever you put\nthe configuration going\nto be reflected in the code.\nFor example, here if you\nchange between the completion\nand response API are going\nto be different really.\nThe completion is really broadly\nused programmatic syntax\nfor submitting\nprompts to a model.\nThe response is\nmore newer syntax\nthat offer greater flexibility\nfor building apps that converse\nwith both standalone models\nand also with agent.\nIf I change the language, so,\nfor example, from Python,\nJavaScript, I can see that\nalso here the code changed a\nlittle bit.\nAlso, if I basically look\ninto the piece of code,\nso when we start here\nfor this with Python,\nI see from OpenAI,\nimport OpenAI.\nSo the first thing we are going\nto import, the libraries.\nAnd then we have two information\nthat we are putting in place,\nthe endpoint and deployment\nname and also the API key.\nRight. The API key, you\ncan change it here.\nFrom three dots, you can change\nfrom the API key to using Entra.\nSo that's something\nthat you can do.\nHere also we are\ninterested to the client.\nSo you can see here for the\nclient I'm using OpenAI,\nand I'm grabbing\nthat information.\nSo the base URL, which\nis the endpoint,\nand the API key I'm providing\nreally the API key, right.\nThen I have here selected\nthe completion API.\nSo it's going to be\nreflected, also dumped,\nwhich is the completion.\nSo here I'm getting the\ncompletion that's going\nto receive the result.\nWe have here I'm using\nthe client, really,\nchat completion create method.\nThat takes the model,\nwhich is the one\nthat configured the\ndeployment name GPT 4.1,\nand the message it composed\nof two things when it comes\nto completion, the\nrole and the content.\nThe content it has,\nyou can see here.\nWhat is the capital of\nFrance, which is the front;\nand the role here I put\nin the role is the user\nthat is asking question.\nIf I change this to really\nresponse API, here's the piece\nof code going to change.\nAnd the difference here is\nthe method that we are using\nand the method how it\nacts with the rest.\nSo we can see here I'm using\nclient response create,\nand I'm putting the\nmobile and the input.\nThen here, if you look\ninto the three dots,\nyou can configure the\nauthentication method\nso we have seen the key.\nNow let's look into the Entra.\nSo, when you use the Entra\nwith the token that you get\nfrom the Entra with the Azure\nidentity, you see first thing\nthat is going to change is the\nlibrary that you load so where\nfrom Azure identity import\ndefault Azure credentials.\nAnd also we have get\ntheir token provider.\nSo this libraries I will\nneed to import when we work\nwith Entra ID and tokens.\nSo here's with Entra ID you're\ngoing to the client app is going\nto represent related\nauthentication and token based\nor an identity that\nis assigned to it.\nSo you can see here I'm\ngetting the token provider.\nI'm using the get bearer token\nprovider using the default\nAzure credentials.\nAnd, when it comes to the\nclient, when I call it,\nthe API key is going to change\nhere to token provider, rather,\nthat we have seen it\npreviously that we provided\nwith the default\nAzure credentials\nand the get bearer credentials.\nThat's only it changed\nwith the connection.\nAnd also I have seen there,\nthere is also there's a link\nto Azure AI that I'm\nshowing in here.\nSo you can configure whatever\nit's more convenient for you,\nso Response API Client\nAPI, the language\nwhen you do your own demo.\nSo you can play around these\ntwo configurations and put it\non whatever is more\nhelpful for you.\nIf you select, for\nexample, here the REST,\nyou can see here we\nhave a curl call.\nAnd, if you stay with OpenAI\nSDK, it's going to be the piece\nof code with the\ncall of OpenAI SDK,\nlike we have\ndiscussed previously.\nLet me go to really Visual\nStudio Code to see this piece\nof code and try to little\nbit play a little bit on it\nand modify it a little bit\nto see how we could possibly\nin real project work on this.\nI'll just set up the\nenvironment now.\nWhen everything is ready here\nand the environment is ready,\nyou can see our workspace.\nWe can see that it shows a\nlittle bit some information\nrelated to other AI Foundry.\nAnd we can really start using\nor trying out our model.\nSo you can see here that\nthe model that I have,\nthe connection to the Azure\ncredential, the token provider,\nthe rule, the content, etc.\nAnd now let's move to Foundry\nto configure or change a\nlittle bit our solution\nto have some more system\nprompt instruction.\nSo now I'm moving it to,\nrather than an AI assistant\nto something more helpful.\nSo here I'm saying\nthat you're going\nto be helpful AI assistant who's\ngoing to support expense claims\nand also provide courses,\naccurate information\nonly on this topic.\nWhen I ask it the\nfirst question,\nwhat kind of business expense\nare typically reimbursed\nby employers; and here's we have\nmore information about this.\nSo some from the web,\ngeneric information,\nanswer about what we can do.\nSo if I ask it, for example,\nsomething out of the scoop,\nfor example, tell me\nabout the ELIZA chatbot,\ngoing to tell me I'm here\nto assist only questions\nrelated to expense claims.\nSo here's a, really, we\nmade our agent specialized\nto specific tasks.\nSo all the change\nthat I made here\nin the code doesn't\nreally appear,\nso the code is just giving you\nan idea of what you can do.\nBut, if I want to have\na specific change,\nI need to change them\nmyself in the code itself.\nOkay. I need to change\nthings in the code itself\nto make things more look like\nmore what I'm expecting.\nSo, for example, here I can\ninclude a section instruction\nthat I will write the same\ninstruction that I put before,\nplus I put the input, which is\nthe prompt, that specific prompt\nthat I want to execute,\nthe user prompt.\nSo, once I'm happy with all\nthis, I can save it as an agent.\nSo just click the button\nsave as an agent,\nand I give it the name.\nI'm going to really name\nit our expense agent.\nSo I'm going to create it\nand name it Expense Agent.\nAnd I just save it to start\nusing it as an agent.\nYeah. Now it's created.\nSo I see it took the\ninstruction that was expected.\nAnd here, interestingly,\npreviously we had\nonly chat and code.\nBut now we do have a\nsection called YAML.\nAnd now let's look into\nthis really YAML section,\nwhat it does contain within it.\nSo, this here, it contains\nreally the definition\nof our agent.\nSo it does include the model,\nthe parameter settings,\nand the instructions that\nwe specified in here.\nSo you can see here\nthis kind of prompt.\nWe have the model GPT 4.1 mini.\nSo, if I look here, for example,\nin this definition here,\nI can see we have the first\nthing we can see here,\nthe kind is prompt, the\nmodel GPT 4.1, instructions,\nthe instruction that we wrote,\nYou are a helpful AI assistant\nof the same restriction that\nwe would route in here.\nWe have also some other\nuseful information\nlike the temperature.\nWe have also the Top P\nand the different tools.\nAnd in this case we are using\nthe web search, basically.\nYou can also look into more\ninformation that it does exist,\nand also you can\nsearch if you want\nto add more in the\ndocumentation.\nHere we have a useful button\nwhere we can do configure\nthe parameters,\nfor example, the temperature.\nWe can see it controls\nthe randomness.\nSo lower interpretation means\nthat the model will\nproduce more repetitive\nand deterministic response;\nor increasing it,\nthe temperature will result\nin more unexpected\nor active responses.\nYou can see also have the top P.\nThis is similar to temperature,\nbut it is controlled\nthe randomness\nbut use different method also.\nSo it depends how\nyou configure it.\nLower on higher are going\nto change the behavior,\nthe general behavior\nof your model.\nYou can see also\nthe text format,\ntext JSON object, JSON schema.\nAnd also you can choose when\nyou use the tools always\nor the model decide\nwhen to choose it.\nSo let's start with\nsome prompts here.\nSo I'm going to ask it,\nbasically, Who are you?\nSo it says, I'm an AI assistant.\nI specialize in employee\nexpenses claim.\nSo it know who it is.\nWe can ask you things like, how\nmuch can I claim for a taxi.\nSo it depends on your\ncompany expense policy.\nThat's what it says.\nSo here we do understand that\nit really needs some data\nto be grounded from\nthe company data.\nSo we need to give it some\ncompany data to be able\nto do that, and for that we\ncan do it through the tools.\nSo here's we can really\nupload some company data\nthrough uploading a file\nthat's going to help\nwith the expense policy.\nSo we're going to\nbrowse for a file.\nSo uploading my file.\nIt's a Word document\nthat has expense policy.\nNow it's going to index it.\nIt has been indexed,\nso it's ready.\nSo this is my file is released\nby having some expense policy\ninformation about taxis, meals,\nentertainment, and so on\nwith the guidelines also.\nSo, once I'm happy with this,\nonce it read really prepared,\ncreated a new index, the\nvector index is there,\nI can attach it and\nstart using it.\nI see here I have\nit file search.\nYou can see it.\nAnd let's see the reflex\nof change what\nhappens to our agent.\nFirst thing, I'm going to\nremove or edit the web search.\nSo I can see here it's\na search Web Bing\nor search specific domain.\nAnd, when I go to the add,\nhere I have the option\nto deactivate the web search.\nNow I go to the YAML file to\nsee the changes that happened,\nso you can see here in the tools\nsection we do have a specific\ntype here, which\nis the web search.\nRight. Once I save,\nthis going to change.\nSo going to reflect here\nwhatever happening here.\nAnd also, in the YAML file,\nI can see the tools we\nhave all the thing.\nIt's not the web anymore.\nIt has a type, which\nis a file search.\nOkay. So here's my\ntools; type file search.\nThey have the vector\nstore for IDs.\nSo it reflected the change.\nSo previously it was a web only,\nand now it reflected the change.\nAlso it's going to reflect it\nwhen we do some\nprompts right now.\nSo let's start new chat.\nGood always; always good\nto start from fresh.\nSo how much can I\nclaim for a taxi?\nNow, reflecting\nlike $500 getting\nfrom that file box\nthat we created.\nAlso, it's going to reflect\nthe change in the tokens\nwhen you upload a file and you\nhave more accurate search.\nSo what about the hotel?\nSo also give me a\nprecise information.\nAlso good option\nthat we have here.\nWe can preview our agents.\nIf I click preview agent, so\nI have like a web browser way\nto interact with my agent.\nSo I can say, How do I submit\nan expense claim, for example.\nI'm going to provide\nthe information.\nAnd this information is\ncoming from the knowledge\nthat we do have in our solution.\nAnd also you can see your code\nwith GitHub, so you have all\nof these options here\nwith the preview.\nSo why not now focusing\non the code part.\nSo you can see here that\nwe do have a code section.\nHere the language\nis focused on C#.\nSo, if you are more C#,\nthis is how it looks in C#.\nWe have also possibility to\nchange it to other languages,\nfor example, Python,\nsince we are talking\nabout Python from earlier.\nSo you can see here the\npiece of code with Python.\nSo here we do have\nfirst delivery.\nSo from Azure identity we grab\nthe default Azure credentials.\nAnd we do have Azure AIProject,\nimport AIProject client.\nAgain, we need the endpoint.\nSo we also, in the\nproject client here,\nwe have the AIProject client\nthat use the endpoint\nand the credentials.\nAnd, for the credentials,\nhere it's using the\ndefault Azure credentials.\nHere we have an agent,\nso my agent is really\nour expense agents.\nThis is the number of\nthe version that I have.\nI see exactly the name of my\nagent, which is express agent.\nAnd here we can see\nfor the response,\nwhich is reference the agent\nto get a response here.\nSo here in the response we see\nthat we are calling OpenAI\nclient response create method,\ntaking the input here,\ntaking the role user, right;\nand the content we are\nsaying, asking the prompt,\ntell me what you can\nhelp with, for example.\nWe also have other information\nlike the agent reference,\nthe name of the agent, also\nthe version of the agent\nthat we are providing\nit to this same method.\nNow let's move to\nVisio Studio Code.\nWe're going to set\nup the environment.\nNow that is ready, we have here\nthe instruction markdown file\nthat show us how to\nwork with this agent.\nSo step by step, open\nterminal, do this.\nAnd, to run your agent locally,\nyou can execute Python\nrun agent command.\nSo I'm going to copy\nthis run agent.py.\nYou can have also\nsee the possibility\nto update your agent\nconfiguration.\nSo you can see here the\nleft-hand activity bar,\nOpen Azure AI Foundry tab.\nUnder Resource, expand\nthe agents section;\nand click corresponding.\nClick Open YAML file,\nmake any changes,\nand then update the changes\nin Azure AI Foundry.\nSo you're really well-guided\nwith this instructions file.\nNow, get you to our agent\nis the same piece of code\nthat we have seen previously\nwith all the information\nthat we read it so far.\nSo the only thing here we should\ndo is really running this agent.\nSo let's start by clicking clear\nto just have a clear terminal.\nGoing to paste the command\nline and see the result.\nSo okay. Here.\nHow to submit an expense claim\ncorrectly, eligibility, etc.,\netc. So it's answering the\nuser prompt that we have here.\nSo we can change it\nto something else.\nFor example, here we\nalready ask questions like,\nhow much can I ask\nclaim for taxes?\nI can take this prompt\nand pass it over here.\nObviously, you can put this as\na parameter of your solution.\nBut here we're trying\njust to make things easy.\nSo here's again\nrunning the agent.\nAnd you see you can claim\nup to $500 per trip.\nProvide me really\ninformation accurate,\ndepending to the\nresource information\nfrom the Word file\nthat we uploaded.\nThat's it for your demo.\nI hope you enjoy it.\nIn this session, we discussed\nhow Microsoft Foundry helps us\ndiscover, compare, and\nsafely use AI models,\nexperiment with them\nusing a playground,\nand ultimately turns them into\nagents that automate tasks\nby combining instructions,\nknowledge, and actions.\nRemember that all this\nmaterial is available for free\non the Learning Path, and you\ncan find the full Learning Path\nat the Microsoft Learn platform.\nThis concludes our session.\nThanks for joining\nus for this session.\nTo continue your learning,\nwe encourage you\nto watch other videos in\nthis course or search\nout your next favorite topic on\nMicrosoft Learn at aka.ms/learn.\nHappy learning.","duration_seconds":2439,"is_auto_generated":false,"transcript_source":"youtube_transcript_api","transcript_status":"available","asr_quality_weight":1.0,"transcript_segments":[{"text":"[ MUSIC ]","start":1.52,"duration":5.82},{"text":"SARAH ALLALI: Okay.","start":7.34,"duration":1.05},{"text":"I want to build an AI assistant,\none that can answer questions,","start":8.39,"duration":5.46},{"text":"check information, and\neven take actions.","start":13.85,"duration":3.94},{"text":"But now I have a\nbigger question.","start":17.79,"duration":2.79},{"text":"Which model should I choose?","start":20.58,"duration":2.34},{"text":"There are so many options,\ndifferent models,","start":22.92,"duration":3.03},{"text":"different capabilities,\ndifferent providers;","start":25.95,"duration":2.76},{"text":"and the choice actually matters","start":28.71,"duration":3.18},{"text":"because the model I choose will\ndirectly impact how my agent","start":31.89,"duration":4.74},{"text":"going to behave.","start":36.63,"duration":2.55},{"text":"So how do I choose\nthe right model","start":39.18,"duration":3.0},{"text":"for the right scenario\ninstead of just guessing?","start":42.18,"duration":3.69},{"text":"What if I had a single place\nwhere I can explore, compare,","start":45.87,"duration":5.01},{"text":"test models before using them?","start":50.88,"duration":3.12},{"text":"That's exactly what\nMicrosoft Foundry comes in.","start":54.0,"duration":3.39},{"text":"If you are wondering how to\nchoose models and use them","start":57.39,"duration":3.33},{"text":"to build a diligent agent,\nyou are in the right place.","start":60.72,"duration":4.08},{"text":"Hi. I'm Sarah Allali, a\nsenior technical trainer","start":64.8,"duration":3.24},{"text":"at Microsoft specializing in AI\nCopilot and building intelligent","start":68.04,"duration":4.8},{"text":"and secure application\non Microsoft Azure.","start":72.84,"duration":3.12},{"text":"I have a PhD in computer science","start":75.96,"duration":2.16},{"text":"and 12 years teaching\ntraining experience.","start":78.12,"duration":3.3},{"text":"This course covers 12\nsessions where we'll dive","start":81.42,"duration":3.39},{"text":"into CoreAI concepts and Azure\nand learn how to turn ideas","start":84.81,"duration":5.07},{"text":"into real applications\nusing modern tools","start":89.88,"duration":3.42},{"text":"like Microsoft Foundry.","start":93.3,"duration":1.83},{"text":"In the first session, we\nwill see how to get started","start":95.13,"duration":3.0},{"text":"with generative AI\nand agents in Azure.","start":98.13,"duration":3.03},{"text":"I will show you how Microsoft\nFoundry helps you discover,","start":101.16,"duration":3.84},{"text":"evaluate, and use AI models","start":105.0,"duration":2.76},{"text":"to build real\napplications and agents.","start":107.76,"duration":3.64},{"text":"Up to now, we have talked\nabout models, agents,","start":112.95,"duration":3.42},{"text":"and how they work together.","start":116.37,"duration":2.55},{"text":"Now let's answer a very\npractical question.","start":118.92,"duration":3.06},{"text":"Where do these models come from,","start":121.98,"duration":2.85},{"text":"and how do we choose\nthe right one?","start":124.83,"duration":2.52},{"text":"That's exactly where\nMicrosoft Foundry comes in.","start":127.35,"duration":4.14},{"text":"Think of Microsoft Foundry\nas a created marketplace","start":131.49,"duration":3.87},{"text":"and delivery platform\nfor AI models.","start":135.36,"duration":2.79},{"text":"Instead of randomly picking\na model and hoping it works,","start":138.15,"duration":4.5},{"text":"Foundry gives you a catalog","start":142.65,"duration":2.7},{"text":"where models are clearly\ndescribed, categorized,","start":145.35,"duration":4.2},{"text":"governed, and delivered in a\nsafe, enterprise-ready way.","start":149.55,"duration":5.28},{"text":"You are not guessing.","start":154.83,"duration":1.08},{"text":"You are choosing intentionally.","start":155.91,"duration":3.27},{"text":"At Microsoft Foundry, we provide\ntwo big model categories,","start":159.18,"duration":4.45},{"text":"models sold directly by\nMicrosoft and models","start":164.76,"duration":3.03},{"text":"from partner and community.","start":167.79,"duration":2.89},{"text":"Models sold directly by\nMicrosoft are models","start":172.02,"duration":3.48},{"text":"that are hosted by\nMicrosoft and built","start":175.5,"duration":2.94},{"text":"through your Azure subscription.","start":178.44,"duration":2.79},{"text":"They come with enterprise\nsupport and SLAs,","start":181.23,"duration":3.84},{"text":"follow Microsoft responsible\nAI standards that we talked","start":185.07,"duration":3.99},{"text":"about in a previous session.","start":189.06,"duration":3.03},{"text":"These include all Azure OpenAI\nmodels and selected models","start":192.09,"duration":3.69},{"text":"from trusted providers that\nMicrosoft manages directly.","start":195.78,"duration":4.83},{"text":"For example, if you are a\ncompany building an internal HR","start":200.61,"duration":3.84},{"text":"assistant and you want\nenterprise security,","start":204.45,"duration":3.45},{"text":"predictable billing,\nand Microsoft support,","start":207.9,"duration":3.18},{"text":"you will typically start with\nAzure OpenAI models in Foundry.","start":211.08,"duration":5.22},{"text":"The partner community models\ngives access to models","start":216.3,"duration":3.54},{"text":"from Research Lab, open\nsource communities,","start":219.84,"duration":3.39},{"text":"and specialized AI providers.","start":223.23,"duration":2.58},{"text":"These models are provided by\ntrusted third-party partners,","start":225.81,"duration":4.77},{"text":"purchased under the partners'\nterms, and are useful for niche","start":230.58,"duration":4.83},{"text":"or specialized use cases.","start":235.41,"duration":3.27},{"text":"For example, you are building an\nAI system for medical imaging","start":238.68,"duration":4.2},{"text":"or legal document analysis.","start":242.88,"duration":2.61},{"text":"A partner or community\nmodel may be a better fit","start":245.49,"duration":3.84},{"text":"than a general purpose\nlanguage model","start":249.33,"duration":1.89},{"text":"for some specific scenarios.","start":251.22,"duration":3.21},{"text":"Foundry doesn't\njust list models.","start":254.43,"duration":2.73},{"text":"It helps you compare\nthem intentionally based","start":257.16,"duration":3.84},{"text":"on real decision factors.","start":261.0,"duration":2.55},{"text":"And let's break those down.","start":263.55,"duration":2.67},{"text":"First, deployment type meta,\nwhere the work happens.","start":266.22,"duration":5.79},{"text":"This answer questions like\nwhere is your data processed?","start":272.01,"duration":3.9},{"text":"How is the model hosted?","start":275.91,"duration":2.04},{"text":"How do you pay for it?","start":277.95,"duration":2.1},{"text":"For example, a public\ncustomer-facing chatbot may use","start":280.05,"duration":4.23},{"text":"a fully managed\ncloud deployment.","start":284.28,"duration":3.33},{"text":"A sensitive internal tool may\nrequire stricter hosting rules.","start":287.61,"duration":6.19},{"text":"The second factor is\nversioning and updates.","start":294.87,"duration":4.98},{"text":"These depends on\nyour preference.","start":299.85,"duration":3.12},{"text":"Do you prefer stability, or\ndo you prefer innovation?","start":302.97,"duration":4.89},{"text":"Models evolve, and Foundry lets\nyou see the model version,","start":307.86,"duration":6.09},{"text":"whether it is auto updated or\nstay fixed for consistency.","start":313.95,"duration":5.82},{"text":"For example, if you are in\na finance or healthcare,","start":319.77,"duration":4.23},{"text":"you may prefer a fixed version.","start":324.0,"duration":2.16},{"text":"If you are experimenting, you\nmay welcome automatic upgrades.","start":326.16,"duration":5.97},{"text":"The third factor is\nabout the rate limits.","start":332.13,"duration":2.94},{"text":"This is all about how\nfast and how often.","start":335.07,"duration":3.57},{"text":"Every model has limits, such as\nthe maximum tokens per minute","start":338.64,"duration":4.71},{"text":"and the throughput constraints.","start":343.35,"duration":2.43},{"text":"So, for example, a high-volume\ncustomer support bots needs","start":345.78,"duration":4.95},{"text":"higher rate limits than a\nsmall internal assistant.","start":350.73,"duration":5.37},{"text":"The last factor is\nabout guardrails","start":356.1,"duration":3.6},{"text":"and responsible AI policies.","start":359.7,"duration":2.4},{"text":"And this is critical.","start":362.1,"duration":1.86},{"text":"Foundry models come with\nbuilt-in responsible AI policy,","start":363.96,"duration":4.71},{"text":"content safety control,\nand usage guidelines.","start":368.67,"duration":4.11},{"text":"For example, you don't want your\nAI assistant generating unsafe","start":372.78,"duration":4.74},{"text":"or inappropriate content.","start":377.52,"duration":2.34},{"text":"Foundry helps apply\nguardrails by design,","start":379.86,"duration":3.9},{"text":"not as an afterthought.","start":383.76,"duration":3.33},{"text":"Now that we understand\nmodels and agents,","start":387.09,"duration":3.27},{"text":"let's look at how we actually\nuse a generative AI model","start":390.36,"duration":3.72},{"text":"in practice.","start":394.08,"duration":1.59},{"text":"And the most important\nrule here is simple:","start":395.67,"duration":3.42},{"text":"Before you write\ncode, you experiment.","start":399.09,"duration":3.75},{"text":"That's exactly why Microsoft\nFoundry provides the","start":402.84,"duration":2.97},{"text":"Model Playground.","start":405.81,"duration":2.64},{"text":"Think of the Model\nPlayground as a sandbox.","start":408.45,"duration":3.18},{"text":"It is a safe place where you\ncan try different models,","start":411.63,"duration":4.53},{"text":"test prompts, adjust settings,\nand see result instantly,","start":416.16,"duration":6.33},{"text":"all without writing a\nsingle line of code.","start":422.49,"duration":3.12},{"text":"For example, you want an AI","start":425.61,"duration":1.56},{"text":"to help employees write\nprofessional emails.","start":427.17,"duration":3.69},{"text":"In the playground, you can type\nor write a polite email asking","start":430.86,"duration":5.43},{"text":"for a deadline extension.","start":436.29,"duration":2.64},{"text":"You see the result immediately.","start":438.93,"duration":2.34},{"text":"Tweak the wording and refine the\ntone before you build anything.","start":441.27,"duration":5.94},{"text":"Why the playground\nis so valuable?","start":447.21,"duration":3.03},{"text":"Because it helps\nyou to test prompt","start":450.24,"duration":2.52},{"text":"by understanding what works\nand what doesn't work.","start":452.76,"duration":4.68},{"text":"Tune parameter by controlling\ntone, length, creativity,","start":457.44,"duration":5.25},{"text":"and also capture configuration","start":462.69,"duration":1.8},{"text":"by saving exact\nsetting before coding.","start":464.49,"duration":4.77},{"text":"This avoids a very common\nmistake, jumping straight","start":469.26,"duration":4.53},{"text":"into code and debugging\nprompt at the same time.","start":473.79,"duration":4.92},{"text":"When you work with a\ngenerative AI model,","start":478.71,"duration":3.0},{"text":"whether in the\nplayground or in code,","start":481.71,"duration":2.34},{"text":"you always configure three\nmain things: instructions,","start":484.05,"duration":4.98},{"text":"input, and parameters.","start":489.03,"duration":2.91},{"text":"Instruction, or system\nprompt, sets the rule","start":491.94,"duration":4.11},{"text":"and the rules for the model.","start":496.05,"duration":3.45},{"text":"Think of it as giving the\nAI a job description.","start":499.5,"duration":3.57},{"text":"For example, for an HR\nagent, we would tell it,","start":503.07,"duration":4.08},{"text":"\"You are a helpful HR assistant.","start":507.15,"duration":3.6},{"text":"You respond professionally\nand clearly.","start":510.75,"duration":3.42},{"text":"You don't give any legal advice.","start":514.17,"duration":4.14},{"text":"This frames how\nthe model behaves.","start":518.31,"duration":3.63},{"text":"Next, we have input,\nor user prompt.","start":521.94,"duration":4.74},{"text":"This is what the\nuser actually asks.","start":526.68,"duration":3.36},{"text":"Good prompts are always clear,\nspecific, and context-rich.","start":530.04,"duration":5.97},{"text":"For example, rather than\nwriting, Write an email,","start":536.01,"duration":5.82},{"text":"in the prompt, you would say,","start":541.83,"duration":2.73},{"text":"Write a friendly email\nreminding a colleague","start":544.56,"duration":3.99},{"text":"about a meeting\ntomorrow at 10am.","start":548.55,"duration":3.66},{"text":"This prompt will get you\na much better result.","start":552.21,"duration":4.56},{"text":"Finally, the parameters.","start":556.77,"duration":1.86},{"text":"The parameters is all about\nhow the model responds.","start":558.63,"duration":4.08},{"text":"Parameters help fine\ntune the response","start":562.71,"duration":4.41},{"text":"by using factors\nsuch as creativity.","start":567.12,"duration":3.68},{"text":"If you put it low, you\nwill get you a precise,","start":570.8,"duration":4.32},{"text":"predictable answers.","start":575.12,"duration":2.19},{"text":"And, if you put it high,","start":577.31,"duration":1.53},{"text":"you will produce a more\ncreative, varied response.","start":578.84,"duration":4.99},{"text":"Another factor is the\nresponse length, like a short","start":583.83,"duration":4.56},{"text":"or maybe a long or maybe\nmore detailed explanation.","start":588.39,"duration":5.4},{"text":"So what is the best\nconfiguration for the factors?","start":593.79,"duration":3.84},{"text":"This depends on the context.","start":597.63,"duration":1.68},{"text":"For example, for\na legal summary,","start":599.31,"duration":2.82},{"text":"low creativity is better;\nand, for marketing copy,","start":602.13,"duration":4.14},{"text":"high creativity works well.","start":606.27,"duration":3.25},{"text":"Once you are happy\nin the playground,","start":610.62,"duration":2.07},{"text":"moving to code is\nstraightforward.","start":612.69,"duration":3.6},{"text":"For example, let's look\nat this without diving","start":616.29,"duration":3.06},{"text":"into technical details.","start":619.35,"duration":1.71},{"text":"In the slide, you have a\ntypical Python example","start":621.06,"duration":3.51},{"text":"that includes an endpoint\nwhere the model lives,","start":624.57,"duration":4.23},{"text":"an API key to\nauthenticate securely,","start":628.8,"duration":3.27},{"text":"and the prompt\nsubmission in the input,","start":632.07,"duration":3.54},{"text":"the same prompt you tested.","start":635.61,"duration":3.87},{"text":"What is important to know the\nexact same concept applied","start":639.48,"duration":3.45},{"text":"in Python, JavaScript,\nC#, or any language.","start":642.93,"duration":3.63},{"text":"Only the syntax is going\nto change, not the logic.","start":646.56,"duration":4.29},{"text":"And yes; you use OpenAI or APIs\nthrough Azure Microsoft Foundry","start":650.85,"duration":5.37},{"text":"to consume these models in code.","start":656.22,"duration":3.51},{"text":"Up to now, we have been working","start":659.73,"duration":2.28},{"text":"with generative AI\nmodels testing prompt,","start":662.01,"duration":3.39},{"text":"tuning parameters,","start":665.4,"duration":1.47},{"text":"and understanding how\nmodels generate responses.","start":666.87,"duration":3.42},{"text":"The natural next step is\nturning a model into an agent.","start":670.29,"duration":4.14},{"text":"That's exactly what the\nMicrosoft Foundry Agent Service","start":674.43,"duration":3.24},{"text":"is for.","start":677.67,"duration":2.37},{"text":"What does creating an\nagent really mean?","start":680.04,"duration":3.48},{"text":"In simple terms, a model answers\nquestions; an agent is designed","start":683.52,"duration":6.3},{"text":"to get something done.","start":689.82,"duration":2.91},{"text":"When you create an\nagent in Foundry,","start":692.73,"duration":2.46},{"text":"you take a model\nconfiguration that works;","start":695.19,"duration":3.45},{"text":"save it as a named agent; give\nit persistent instruction; and,","start":698.64,"duration":5.94},{"text":"finally, connect it to\ntools and knowledge","start":704.58,"duration":4.05},{"text":"so the agent doesn't forget its\nrole every time you talk to it.","start":708.63,"duration":5.7},{"text":"One key difference from models\nis that agents remember","start":714.33,"duration":4.08},{"text":"who they are and what\nthey are supposed to do.","start":718.41,"duration":4.08},{"text":"Their instructions act\nlike a job description.","start":722.49,"duration":4.83},{"text":"What they help with,\nwhat tool to use,","start":727.32,"duration":2.97},{"text":"and what they should not do,\nthis makes agent consistent","start":730.29,"duration":4.59}],"view_count_at_fetch":86,"transcript_available":true,"source_universe_backfill":{"version":"source_universe_backfill.v1","backfilled_at":"2026-07-01T06:05:11.014475+00:00","resolver_warnings":[],"resolved_source_family":"video_media"},"transcript_last_attempted_at":"2026-07-01T06:08:15.433652+00:00"},"tags_json":[],"is_discovered_source":false},"transcript":{"segment_count":200,"markdown":"# Transcript\n\n## Segment 1\n\n**Speaker:** Unknown speaker\n\n[ MUSIC ]\n\n## Segment 2\n\n**Speaker:** Unknown speaker\n\nSARAH ALLALI: Okay.\n\n## Segment 3\n\n**Speaker:** Unknown speaker\n\nI want to build an AI assistant, one that can answer questions,\n\n## Segment 4\n\n**Speaker:** Unknown speaker\n\ncheck information, and even take actions.\n\n## Segment 5\n\n**Speaker:** Unknown speaker\n\nBut now I have a bigger question.\n\n## Segment 6\n\n**Speaker:** Unknown speaker\n\nWhich model should I choose?\n\n## Segment 7\n\n**Speaker:** Unknown speaker\n\nThere are so many options, different models,\n\n## Segment 8\n\n**Speaker:** Unknown speaker\n\ndifferent capabilities, different providers;\n\n## Segment 9\n\n**Speaker:** Unknown speaker\n\nand the choice actually matters\n\n## Segment 10\n\n**Speaker:** Unknown speaker\n\nbecause the model I choose will directly impact how my agent\n\n## Segment 11\n\n**Speaker:** Unknown speaker\n\ngoing to behave.\n\n## Segment 12\n\n**Speaker:** Unknown speaker\n\nSo how do I choose the right model\n\n## Segment 13\n\n**Speaker:** Unknown speaker\n\nfor the right scenario instead of just guessing?\n\n## Segment 14\n\n**Speaker:** Unknown speaker\n\nWhat if I had a single place where I can explore, compare,\n\n## Segment 15\n\n**Speaker:** Unknown speaker\n\ntest models before using them?\n\n## Segment 16\n\n**Speaker:** Unknown speaker\n\nThat's exactly what Microsoft Foundry comes in.\n\n## Segment 17\n\n**Speaker:** Unknown speaker\n\nIf you are wondering how to choose models and use them\n\n## Segment 18\n\n**Speaker:** Unknown speaker\n\nto build a diligent agent, you are in the right place.\n\n## Segment 19\n\n**Speaker:** Unknown speaker\n\nHi. I'm Sarah Allali, a senior technical trainer\n\n## Segment 20\n\n**Speaker:** Unknown speaker\n\nat Microsoft specializing in AI Copilot and building intelligent\n\n## Segment 21\n\n**Speaker:** Unknown speaker\n\nand secure application on Microsoft Azure.\n\n## Segment 22\n\n**Speaker:** Unknown speaker\n\nI have a PhD in computer science\n\n## Segment 23\n\n**Speaker:** Unknown speaker\n\nand 12 years teaching training experience.\n\n## Segment 24\n\n**Speaker:** Unknown speaker\n\nThis course covers 12 sessions where we'll dive\n\n## Segment 25\n\n**Speaker:** Unknown speaker\n\ninto CoreAI concepts and Azure and learn how to turn ideas\n\n## Segment 26\n\n**Speaker:** Unknown speaker\n\ninto real applications using modern tools\n\n## Segment 27\n\n**Speaker:** Unknown speaker\n\nlike Microsoft Foundry.\n\n## Segment 28\n\n**Speaker:** Unknown speaker\n\nIn the first session, we will see how to get started\n\n## Segment 29\n\n**Speaker:** Unknown speaker\n\nwith generative AI and agents in Azure.\n\n## Segment 30\n\n**Speaker:** Unknown speaker\n\nI will show you how Microsoft Foundry helps you discover,\n\n## Segment 31\n\n**Speaker:** Unknown speaker\n\nevaluate, and use AI models\n\n## Segment 32\n\n**Speaker:** Unknown speaker\n\nto build real applications and agents.\n\n## Segment 33\n\n**Speaker:** Unknown speaker\n\nUp to now, we have talked about models, agents,\n\n## Segment 34\n\n**Speaker:** Unknown speaker\n\nand how they work together.\n\n## Segment 35\n\n**Speaker:** Unknown speaker\n\nNow let's answer a very practical question.\n\n## Segment 36\n\n**Speaker:** Unknown speaker\n\nWhere do these models come from,\n\n## Segment 37\n\n**Speaker:** Unknown speaker\n\nand how do we choose the right one?\n\n## Segment 38\n\n**Speaker:** Unknown speaker\n\nThat's exactly where Microsoft Foundry comes in.\n\n## Segment 39\n\n**Speaker:** Unknown speaker\n\nThink of Microsoft Foundry as a created marketplace\n\n## Segment 40\n\n**Speaker:** Unknown speaker\n\nand delivery platform for AI models.\n\n## Segment 41\n\n**Speaker:** Unknown speaker\n\nInstead of randomly picking a model and hoping it works,\n\n## Segment 42\n\n**Speaker:** Unknown speaker\n\nFoundry gives you a catalog\n\n## Segment 43\n\n**Speaker:** Unknown speaker\n\nwhere models are clearly described, categorized,\n\n## Segment 44\n\n**Speaker:** Unknown speaker\n\ngoverned, and delivered in a safe, enterprise-ready way.\n\n## Segment 45\n\n**Speaker:** Unknown speaker\n\nYou are not guessing.\n\n## Segment 46\n\n**Speaker:** Unknown speaker\n\nYou are choosing intentionally.\n\n## Segment 47\n\n**Speaker:** Unknown speaker\n\nAt Microsoft Foundry, we provide two big model categories,\n\n## Segment 48\n\n**Speaker:** Unknown speaker\n\nmodels sold directly by Microsoft and models\n\n## Segment 49\n\n**Speaker:** Unknown speaker\n\nfrom partner and community.\n\n## Segment 50\n\n**Speaker:** Unknown speaker\n\nModels sold directly by Microsoft are models\n\n## Segment 51\n\n**Speaker:** Unknown speaker\n\nthat are hosted by Microsoft and built\n\n## Segment 52\n\n**Speaker:** Unknown speaker\n\nthrough your Azure subscription.\n\n## Segment 53\n\n**Speaker:** Unknown speaker\n\nThey come with enterprise support and SLAs,\n\n## Segment 54\n\n**Speaker:** Unknown speaker\n\nfollow Microsoft responsible AI standards that we talked\n\n## Segment 55\n\n**Speaker:** Unknown speaker\n\nabout in a previous session.\n\n## Segment 56\n\n**Speaker:** Unknown speaker\n\nThese include all Azure OpenAI models and selected models\n\n## Segment 57\n\n**Speaker:** Unknown speaker\n\nfrom trusted providers that Microsoft manages directly.\n\n## Segment 58\n\n**Speaker:** Unknown speaker\n\nFor example, if you are a company building an internal HR\n\n## Segment 59\n\n**Speaker:** Unknown speaker\n\nassistant and you want enterprise security,\n\n## Segment 60\n\n**Speaker:** Unknown speaker\n\npredictable billing, and Microsoft support,\n\n## Segment 61\n\n**Speaker:** Unknown speaker\n\nyou will typically start with Azure OpenAI models in Foundry.\n\n## Segment 62\n\n**Speaker:** Unknown speaker\n\nThe partner community models gives access to models\n\n## Segment 63\n\n**Speaker:** Unknown speaker\n\nfrom Research Lab, open source communities,\n\n## Segment 64\n\n**Speaker:** Unknown speaker\n\nand specialized AI providers.\n\n## Segment 65\n\n**Speaker:** Unknown speaker\n\nThese models are provided by trusted third-party partners,\n\n## Segment 66\n\n**Speaker:** Unknown speaker\n\npurchased under the partners' terms, and are useful for niche\n\n## Segment 67\n\n**Speaker:** Unknown speaker\n\nor specialized use cases.\n\n## Segment 68\n\n**Speaker:** Unknown speaker\n\nFor example, you are building an AI system for medical imaging\n\n## Segment 69\n\n**Speaker:** Unknown speaker\n\nor legal document analysis.\n\n## Segment 70\n\n**Speaker:** Unknown speaker\n\nA partner or community model may be a better fit\n\n## Segment 71\n\n**Speaker:** Unknown speaker\n\nthan a general purpose language model\n\n## Segment 72\n\n**Speaker:** Unknown speaker\n\nfor some specific scenarios.\n\n## Segment 73\n\n**Speaker:** Unknown speaker\n\nFoundry doesn't just list models.\n\n## Segment 74\n\n**Speaker:** Unknown speaker\n\nIt helps you compare them intentionally based\n\n## Segment 75\n\n**Speaker:** Unknown speaker\n\non real decision factors.\n\n## Segment 76\n\n**Speaker:** Unknown speaker\n\nAnd let's break those down.\n\n## Segment 77\n\n**Speaker:** Unknown speaker\n\nFirst, deployment type meta, where the work happens.\n\n## Segment 78\n\n**Speaker:** Unknown speaker\n\nThis answer questions like where is your data processed?\n\n## Segment 79\n\n**Speaker:** Unknown speaker\n\nHow is the model hosted?\n\n## Segment 80\n\n**Speaker:** Unknown speaker\n\nHow do you pay for it?\n\n## Segment 81\n\n**Speaker:** Unknown speaker\n\nFor example, a public customer-facing chatbot may use\n\n## Segment 82\n\n**Speaker:** Unknown speaker\n\na fully managed cloud deployment.\n\n## Segment 83\n\n**Speaker:** Unknown speaker\n\nA sensitive internal tool may require stricter hosting rules.\n\n## Segment 84\n\n**Speaker:** Unknown speaker\n\nThe second factor is versioning and updates.\n\n## Segment 85\n\n**Speaker:** Unknown speaker\n\nThese depends on your preference.\n\n## Segment 86\n\n**Speaker:** Unknown speaker\n\nDo you prefer stability, or do you prefer innovation?\n\n## Segment 87\n\n**Speaker:** Unknown speaker\n\nModels evolve, and Foundry lets you see the model version,\n\n## Segment 88\n\n**Speaker:** Unknown speaker\n\nwhether it is auto updated or stay fixed for consistency.\n\n## Segment 89\n\n**Speaker:** Unknown speaker\n\nFor example, if you are in a finance or healthcare,\n\n## Segment 90\n\n**Speaker:** Unknown speaker\n\nyou may prefer a fixed version.\n\n## Segment 91\n\n**Speaker:** Unknown speaker\n\nIf you are experimenting, you may welcome automatic upgrades.\n\n## Segment 92\n\n**Speaker:** Unknown speaker\n\nThe third factor is about the rate limits.\n\n## Segment 93\n\n**Speaker:** Unknown speaker\n\nThis is all about how fast and how often.\n\n## Segment 94\n\n**Speaker:** Unknown speaker\n\nEvery model has limits, such as the maximum tokens per minute\n\n## Segment 95\n\n**Speaker:** Unknown speaker\n\nand the throughput constraints.\n\n## Segment 96\n\n**Speaker:** Unknown speaker\n\nSo, for example, a high-volume customer support bots needs\n\n## Segment 97\n\n**Speaker:** Unknown speaker\n\nhigher rate limits than a small internal assistant.\n\n## Segment 98\n\n**Speaker:** Unknown speaker\n\nThe last factor is about guardrails\n\n## Segment 99\n\n**Speaker:** Unknown speaker\n\nand responsible AI policies.\n\n## Segment 100\n\n**Speaker:** Unknown speaker\n\nAnd this is critical.\n\n## Segment 101\n\n**Speaker:** Unknown speaker\n\nFoundry models come with built-in responsible AI policy,\n\n## Segment 102\n\n**Speaker:** Unknown speaker\n\ncontent safety control, and usage guidelines.\n\n## Segment 103\n\n**Speaker:** Unknown speaker\n\nFor example, you don't want your AI assistant generating unsafe\n\n## Segment 104\n\n**Speaker:** Unknown speaker\n\nor inappropriate content.\n\n## Segment 105\n\n**Speaker:** Unknown speaker\n\nFoundry helps apply guardrails by design,\n\n## Segment 106\n\n**Speaker:** Unknown speaker\n\nnot as an afterthought.\n\n## Segment 107\n\n**Speaker:** Unknown speaker\n\nNow that we understand models and agents,\n\n## Segment 108\n\n**Speaker:** Unknown speaker\n\nlet's look at how we actually use a generative AI model\n\n## Segment 109\n\n**Speaker:** Unknown speaker\n\nin practice.\n\n## Segment 110\n\n**Speaker:** Unknown speaker\n\nAnd the most important rule here is simple:\n\n## Segment 111\n\n**Speaker:** Unknown speaker\n\nBefore you write code, you experiment.\n\n## Segment 112\n\n**Speaker:** Unknown speaker\n\nThat's exactly why Microsoft Foundry provides the\n\n## Segment 113\n\n**Speaker:** Unknown speaker\n\nModel Playground.\n\n## Segment 114\n\n**Speaker:** Unknown speaker\n\nThink of the Model Playground as a sandbox.\n\n## Segment 115\n\n**Speaker:** Unknown speaker\n\nIt is a safe place where you can try different models,\n\n## Segment 116\n\n**Speaker:** Unknown speaker\n\ntest prompts, adjust settings, and see result instantly,\n\n## Segment 117\n\n**Speaker:** Unknown speaker\n\nall without writing a single line of code.\n\n## Segment 118\n\n**Speaker:** Unknown speaker\n\nFor example, you want an AI\n\n## Segment 119\n\n**Speaker:** Unknown speaker\n\nto help employees write professional emails.\n\n## Segment 120\n\n**Speaker:** Unknown speaker\n\nIn the playground, you can type or write a polite email asking\n\n## Segment 121\n\n**Speaker:** Unknown speaker\n\nfor a deadline extension.\n\n## Segment 122\n\n**Speaker:** Unknown speaker\n\nYou see the result immediately.\n\n## Segment 123\n\n**Speaker:** Unknown speaker\n\nTweak the wording and refine the tone before you build anything.\n\n## Segment 124\n\n**Speaker:** Unknown speaker\n\nWhy the playground is so valuable?\n\n## Segment 125\n\n**Speaker:** Unknown speaker\n\nBecause it helps you to test prompt\n\n## Segment 126\n\n**Speaker:** Unknown speaker\n\nby understanding what works and what doesn't work.\n\n## Segment 127\n\n**Speaker:** Unknown speaker\n\nTune parameter by controlling tone, length, creativity,\n\n## Segment 128\n\n**Speaker:** Unknown speaker\n\nand also capture configuration\n\n## Segment 129\n\n**Speaker:** Unknown speaker\n\nby saving exact setting before coding.\n\n## Segment 130\n\n**Speaker:** Unknown speaker\n\nThis avoids a very common mistake, jumping straight\n\n## Segment 131\n\n**Speaker:** Unknown speaker\n\ninto code and debugging prompt at the same time.\n\n## Segment 132\n\n**Speaker:** Unknown speaker\n\nWhen you work with a generative AI model,\n\n## Segment 133\n\n**Speaker:** Unknown speaker\n\nwhether in the playground or in code,\n\n## Segment 134\n\n**Speaker:** Unknown speaker\n\nyou always configure three main things: instructions,\n\n## Segment 135\n\n**Speaker:** Unknown speaker\n\ninput, and parameters.\n\n## Segment 136\n\n**Speaker:** Unknown speaker\n\nInstruction, or system prompt, sets the rule\n\n## Segment 137\n\n**Speaker:** Unknown speaker\n\nand the rules for the model.\n\n## Segment 138\n\n**Speaker:** Unknown speaker\n\nThink of it as giving the AI a job description.\n\n## Segment 139\n\n**Speaker:** Unknown speaker\n\nFor example, for an HR agent, we would tell it,\n\n## Segment 140\n\n**Speaker:** Unknown speaker\n\n\"You are a helpful HR assistant.\n\n## Segment 141\n\n**Speaker:** Unknown speaker\n\nYou respond professionally and clearly.\n\n## Segment 142\n\n**Speaker:** Unknown speaker\n\nYou don't give any legal advice.\n\n## Segment 143\n\n**Speaker:** Unknown speaker\n\nThis frames how the model behaves.\n\n## Segment 144\n\n**Speaker:** Unknown speaker\n\nNext, we have input, or user prompt.\n\n## Segment 145\n\n**Speaker:** Unknown speaker\n\nThis is what the user actually asks.\n\n## Segment 146\n\n**Speaker:** Unknown speaker\n\nGood prompts are always clear, specific, and context-rich.\n\n## Segment 147\n\n**Speaker:** Unknown speaker\n\nFor example, rather than writing, Write an email,\n\n## Segment 148\n\n**Speaker:** Unknown speaker\n\nin the prompt, you would say,\n\n## Segment 149\n\n**Speaker:** Unknown speaker\n\nWrite a friendly email reminding a colleague\n\n## Segment 150\n\n**Speaker:** Unknown speaker\n\nabout a meeting tomorrow at 10am.\n\n## Segment 151\n\n**Speaker:** Unknown speaker\n\nThis prompt will get you a much better result.\n\n## Segment 152\n\n**Speaker:** Unknown speaker\n\nFinally, the parameters.\n\n## Segment 153\n\n**Speaker:** Unknown speaker\n\nThe parameters is all about how the model responds.\n\n## Segment 154\n\n**Speaker:** Unknown speaker\n\nParameters help fine tune the response\n\n## Segment 155\n\n**Speaker:** Unknown speaker\n\nby using factors such as creativity.\n\n## Segment 156\n\n**Speaker:** Unknown speaker\n\nIf you put it low, you will get you a precise,\n\n## Segment 157\n\n**Speaker:** Unknown speaker\n\npredictable answers.\n\n## Segment 158\n\n**Speaker:** Unknown speaker\n\nAnd, if you put it high,\n\n## Segment 159\n\n**Speaker:** Unknown speaker\n\nyou will produce a more creative, varied response.\n\n## Segment 160\n\n**Speaker:** Unknown speaker\n\nAnother factor is the response length, like a short\n\n## Segment 161\n\n**Speaker:** Unknown speaker\n\nor maybe a long or maybe more detailed explanation.\n\n## Segment 162\n\n**Speaker:** Unknown speaker\n\nSo what is the best configuration for the factors?\n\n## Segment 163\n\n**Speaker:** Unknown speaker\n\nThis depends on the context.\n\n## Segment 164\n\n**Speaker:** Unknown speaker\n\nFor example, for a legal summary,\n\n## Segment 165\n\n**Speaker:** Unknown speaker\n\nlow creativity is better; and, for marketing copy,\n\n## Segment 166\n\n**Speaker:** Unknown speaker\n\nhigh creativity works well.\n\n## Segment 167\n\n**Speaker:** Unknown speaker\n\nOnce you are happy in the playground,\n\n## Segment 168\n\n**Speaker:** Unknown speaker\n\nmoving to code is straightforward.\n\n## Segment 169\n\n**Speaker:** Unknown speaker\n\nFor example, let's look at this without diving\n\n## Segment 170\n\n**Speaker:** Unknown speaker\n\ninto technical details.\n\n## Segment 171\n\n**Speaker:** Unknown speaker\n\nIn the slide, you have a typical Python example\n\n## Segment 172\n\n**Speaker:** Unknown speaker\n\nthat includes an endpoint where the model lives,\n\n## Segment 173\n\n**Speaker:** Unknown speaker\n\nan API key to authenticate securely,\n\n## Segment 174\n\n**Speaker:** Unknown speaker\n\nand the prompt submission in the input,\n\n## Segment 175\n\n**Speaker:** Unknown speaker\n\nthe same prompt you tested.\n\n## Segment 176\n\n**Speaker:** Unknown speaker\n\nWhat is important to know the exact same concept applied\n\n## Segment 177\n\n**Speaker:** Unknown speaker\n\nin Python, JavaScript, C#, or any language.\n\n## Segment 178\n\n**Speaker:** Unknown speaker\n\nOnly the syntax is going to change, not the logic.\n\n## Segment 179\n\n**Speaker:** Unknown speaker\n\nAnd yes; you use OpenAI or APIs through Azure Microsoft Foundry\n\n## Segment 180\n\n**Speaker:** Unknown speaker\n\nto consume these models in code.\n\n## Segment 181\n\n**Speaker:** Unknown speaker\n\nUp to now, we have been working\n\n## Segment 182\n\n**Speaker:** Unknown speaker\n\nwith generative AI models testing prompt,\n\n## Segment 183\n\n**Speaker:** Unknown speaker\n\ntuning parameters,\n\n## Segment 184\n\n**Speaker:** Unknown speaker\n\nand understanding how models generate responses.\n\n## Segment 185\n\n**Speaker:** Unknown speaker\n\nThe natural next step is turning a model into an agent.\n\n## Segment 186\n\n**Speaker:** Unknown speaker\n\nThat's exactly what the Microsoft Foundry Agent Service\n\n## Segment 187\n\n**Speaker:** Unknown speaker\n\nis for.\n\n## Segment 188\n\n**Speaker:** Unknown speaker\n\nWhat does creating an agent really mean?\n\n## Segment 189\n\n**Speaker:** Unknown speaker\n\nIn simple terms, a model answers questions; an agent is designed\n\n## Segment 190\n\n**Speaker:** Unknown speaker\n\nto get something done.\n\n## Segment 191\n\n**Speaker:** Unknown speaker\n\nWhen you create an agent in Foundry,\n\n## Segment 192\n\n**Speaker:** Unknown speaker\n\nyou take a model configuration that works;\n\n## Segment 193\n\n**Speaker:** Unknown speaker\n\nsave it as a named agent; give it persistent instruction; and,\n\n## Segment 194\n\n**Speaker:** Unknown speaker\n\nfinally, connect it to tools and knowledge\n\n## Segment 195\n\n**Speaker:** Unknown speaker\n\nso the agent doesn't forget its role every time you talk to it.\n\n## Segment 196\n\n**Speaker:** Unknown speaker\n\nOne key difference from models is that agents remember\n\n## Segment 197\n\n**Speaker:** Unknown speaker\n\nwho they are and what they are supposed to do.\n\n## Segment 198\n\n**Speaker:** Unknown speaker\n\nTheir instructions act like a job description.\n\n## Segment 199\n\n**Speaker:** Unknown speaker\n\nWhat they help with, what tool to use,\n\n## Segment 200\n\n**Speaker:** Unknown speaker\n\nand what they should not do, this makes agent consistent","text":"[segment 0] Unknown speaker: [ MUSIC ]\n[segment 1] Unknown speaker: SARAH ALLALI: Okay.\n[segment 2] Unknown speaker: I want to build an AI assistant, one that can answer questions,\n[segment 3] Unknown speaker: check information, and even take actions.\n[segment 4] Unknown speaker: But now I have a bigger question.\n[segment 5] Unknown speaker: Which model should I choose?\n[segment 6] Unknown speaker: There are so many options, different models,\n[segment 7] Unknown speaker: different capabilities, different providers;\n[segment 8] Unknown speaker: and the choice actually matters\n[segment 9] Unknown speaker: because the model I choose will directly impact how my agent\n[segment 10] Unknown speaker: going to behave.\n[segment 11] Unknown speaker: So how do I choose the right model\n[segment 12] Unknown speaker: for the right scenario instead of just guessing?\n[segment 13] Unknown speaker: What if I had a single place where I can explore, compare,\n[segment 14] Unknown speaker: test models before using them?\n[segment 15] Unknown speaker: That's exactly what Microsoft Foundry comes in.\n[segment 16] Unknown speaker: If you are wondering how to choose models and use them\n[segment 17] Unknown speaker: to build a diligent agent, you are in the right place.\n[segment 18] Unknown speaker: Hi. I'm Sarah Allali, a senior technical trainer\n[segment 19] Unknown speaker: at Microsoft specializing in AI Copilot and building intelligent\n[segment 20] Unknown speaker: and secure application on Microsoft Azure.\n[segment 21] Unknown speaker: I have a PhD in computer science\n[segment 22] Unknown speaker: and 12 years teaching training experience.\n[segment 23] Unknown speaker: This course covers 12 sessions where we'll dive\n[segment 24] Unknown speaker: into CoreAI concepts and Azure and learn how to turn ideas\n[segment 25] Unknown speaker: into real applications using modern tools\n[segment 26] Unknown speaker: like Microsoft Foundry.\n[segment 27] Unknown speaker: In the first session, we will see how to get started\n[segment 28] Unknown speaker: with generative AI and agents in Azure.\n[segment 29] Unknown speaker: I will show you how Microsoft Foundry helps you discover,\n[segment 30] Unknown speaker: evaluate, and use AI models\n[segment 31] Unknown speaker: to build real applications and agents.\n[segment 32] Unknown speaker: Up to now, we have talked about models, agents,\n[segment 33] Unknown speaker: and how they work together.\n[segment 34] Unknown speaker: Now let's answer a very practical question.\n[segment 35] Unknown speaker: Where do these models come from,\n[segment 36] Unknown speaker: and how do we choose the right one?\n[segment 37] Unknown speaker: That's exactly where Microsoft Foundry comes in.\n[segment 38] Unknown speaker: Think of Microsoft Foundry as a created marketplace\n[segment 39] Unknown speaker: and delivery platform for AI models.\n[segment 40] Unknown speaker: Instead of randomly picking a model and hoping it works,\n[segment 41] Unknown speaker: Foundry gives you a catalog\n[segment 42] Unknown speaker: where models are clearly described, categorized,\n[segment 43] Unknown speaker: governed, and delivered in a safe, enterprise-ready way.\n[segment 44] Unknown speaker: You are not guessing.\n[segment 45] Unknown speaker: You are choosing intentionally.\n[segment 46] Unknown speaker: At Microsoft Foundry, we provide two big model categories,\n[segment 47] Unknown speaker: models sold directly by Microsoft and models\n[segment 48] Unknown speaker: from partner and community.\n[segment 49] Unknown speaker: Models sold directly by Microsoft are models\n[segment 50] Unknown speaker: that are hosted by Microsoft and built\n[segment 51] Unknown speaker: through your Azure subscription.\n[segment 52] Unknown speaker: They come with enterprise support and SLAs,\n[segment 53] Unknown speaker: follow Microsoft responsible AI standards that we talked\n[segment 54] Unknown speaker: about in a previous session.\n[segment 55] Unknown speaker: These include all Azure OpenAI models and selected models\n[segment 56] Unknown speaker: from trusted providers that Microsoft manages directly.\n[segment 57] Unknown speaker: For example, if you are a company building an internal HR\n[segment 58] Unknown speaker: assistant and you want enterprise security,\n[segment 59] Unknown speaker: predictable billing, and Microsoft support,\n[segment 60] Unknown speaker: you will typically start with Azure OpenAI models in Foundry.\n[segment 61] Unknown speaker: The partner community models gives access to models\n[segment 62] Unknown speaker: from Research Lab, open source communities,\n[segment 63] Unknown speaker: and specialized AI providers.\n[segment 64] Unknown speaker: These models are provided by trusted third-party partners,\n[segment 65] Unknown speaker: purchased under the partners' terms, and are useful for niche\n[segment 66] Unknown speaker: or specialized use cases.\n[segment 67] Unknown speaker: For example, you are building an AI system for medical imaging\n[segment 68] Unknown speaker: or legal document analysis.\n[segment 69] Unknown speaker: A partner or community model may be a better fit\n[segment 70] Unknown speaker: than a general purpose language model\n[segment 71] Unknown speaker: for some specific scenarios.\n[segment 72] Unknown speaker: Foundry doesn't just list models.\n[segment 73] Unknown speaker: It helps you compare them intentionally based\n[segment 74] Unknown speaker: on real decision factors.\n[segment 75] Unknown speaker: And let's break those down.\n[segment 76] Unknown speaker: First, deployment type meta, where the work happens.\n[segment 77] Unknown speaker: This answer questions like where is your data processed?\n[segment 78] Unknown speaker: How is the model hosted?\n[segment 79] Unknown speaker: How do you pay for it?\n[segment 80] Unknown speaker: For example, a public customer-facing chatbot may use\n[segment 81] Unknown speaker: a fully managed cloud deployment.\n[segment 82] Unknown speaker: A sensitive internal tool may require stricter hosting rules.\n[segment 83] Unknown speaker: The second factor is versioning and updates.\n[segment 84] Unknown speaker: These depends on your preference.\n[segment 85] Unknown speaker: Do you prefer stability, or do you prefer innovation?\n[segment 86] Unknown speaker: Models evolve, and Foundry lets you see the model version,\n[segment 87] Unknown speaker: whether it is auto updated or stay fixed for consistency.\n[segment 88] Unknown speaker: For example, if you are in a finance or healthcare,\n[segment 89] Unknown speaker: you may prefer a fixed version.\n[segment 90] Unknown speaker: If you are experimenting, you may welcome automatic upgrades.\n[segment 91] Unknown speaker: The third factor is about the rate limits.\n[segment 92] Unknown speaker: This is all about how fast and how often.\n[segment 93] Unknown speaker: Every model has limits, such as the maximum tokens per minute\n[segment 94] Unknown speaker: and the throughput constraints.\n[segment 95] Unknown speaker: So, for example, a high-volume customer support bots needs\n[segment 96] Unknown speaker: higher rate limits than a small internal assistant.\n[segment 97] Unknown speaker: The last factor is about guardrails\n[segment 98] Unknown speaker: and responsible AI policies.\n[segment 99] Unknown speaker: And this is critical.\n[segment 100] Unknown speaker: Foundry models come with built-in responsible AI policy,\n[segment 101] Unknown speaker: content safety control, and usage guidelines.\n[segment 102] Unknown speaker: For example, you don't want your AI assistant generating unsafe\n[segment 103] Unknown speaker: or inappropriate content.\n[segment 104] Unknown speaker: Foundry helps apply guardrails by design,\n[segment 105] Unknown speaker: not as an afterthought.\n[segment 106] Unknown speaker: Now that we understand models and agents,\n[segment 107] Unknown speaker: let's look at how we actually use a generative AI model\n[segment 108] Unknown speaker: in practice.\n[segment 109] Unknown speaker: And the most important rule here is simple:\n[segment 110] Unknown speaker: Before you write code, you experiment.\n[segment 111] Unknown speaker: That's exactly why Microsoft Foundry provides the\n[segment 112] Unknown speaker: Model Playground.\n[segment 113] Unknown speaker: Think of the Model Playground as a sandbox.\n[segment 114] Unknown speaker: It is a safe place where you can try different models,\n[segment 115] Unknown speaker: test prompts, adjust settings, and see result instantly,\n[segment 116] Unknown speaker: all without writing a single line of code.\n[segment 117] Unknown speaker: For example, you want an AI\n[segment 118] Unknown speaker: to help employees write professional emails.\n[segment 119] Unknown speaker: In the playground, you can type or write a polite email asking\n[segment 120] Unknown speaker: for a deadline extension.\n[segment 121] Unknown speaker: You see the result immediately.\n[segment 122] Unknown speaker: Tweak the wording and refine the tone before you build anything.\n[segment 123] Unknown speaker: Why the playground is so valuable?\n[segment 124] Unknown speaker: Because it helps you to test prompt\n[segment 125] Unknown speaker: by understanding what works and what doesn't work.\n[segment 126] Unknown speaker: Tune parameter by controlling tone, length, creativity,\n[segment 127] Unknown speaker: and also capture configuration\n[segment 128] Unknown speaker: by saving exact setting before coding.\n[segment 129] Unknown speaker: This avoids a very common mistake, jumping straight\n[segment 130] Unknown speaker: into code and debugging prompt at the same time.\n[segment 131] Unknown speaker: When you work with a generative AI model,\n[segment 132] Unknown speaker: whether in the playground or in code,\n[segment 133] Unknown speaker: you always configure three main things: instructions,\n[segment 134] Unknown speaker: input, and parameters.\n[segment 135] Unknown speaker: Instruction, or system prompt, sets the rule\n[segment 136] Unknown speaker: and the rules for the model.\n[segment 137] Unknown speaker: Think of it as giving the AI a job description.\n[segment 138] Unknown speaker: For example, for an HR agent, we would tell it,\n[segment 139] Unknown speaker: \"You are a helpful HR assistant.\n[segment 140] Unknown speaker: You respond professionally and clearly.\n[segment 141] Unknown speaker: You don't give any legal advice.\n[segment 142] Unknown speaker: This frames how the model behaves.\n[segment 143] Unknown speaker: Next, we have input, or user prompt.\n[segment 144] Unknown speaker: This is what the user actually asks.\n[segment 145] Unknown speaker: Good prompts are always clear, specific, and context-rich.\n[segment 146] Unknown speaker: For example, rather than writing, Write an email,\n[segment 147] Unknown speaker: in the prompt, you would say,\n[segment 148] Unknown speaker: Write a friendly email reminding a colleague\n[segment 149] Unknown speaker: about a meeting tomorrow at 10am.\n[segment 150] Unknown speaker: This prompt will get you a much better result.\n[segment 151] Unknown speaker: Finally, the parameters.\n[segment 152] Unknown speaker: The parameters is all about how the model responds.\n[segment 153] Unknown speaker: Parameters help fine tune the response\n[segment 154] Unknown speaker: by using factors such as creativity.\n[segment 155] Unknown speaker: If you put it low, you will get you a precise,\n[segment 156] Unknown speaker: predictable answers.\n[segment 157] Unknown speaker: And, if you put it high,\n[segment 158] Unknown speaker: you will produce a more creative, varied response.\n[segment 159] Unknown speaker: Another factor is the response length, like a short\n[segment 160] Unknown speaker: or maybe a long or maybe more detailed explanation.\n[segment 161] Unknown speaker: So what is the best configuration for the factors?\n[segment 162] Unknown speaker: This depends on the context.\n[segment 163] Unknown speaker: For example, for a legal summary,\n[segment 164] Unknown speaker: low creativity is better; and, for marketing copy,\n[segment 165] Unknown speaker: high creativity works well.\n[segment 166] Unknown speaker: Once you are happy in the playground,\n[segment 167] Unknown speaker: moving to code is straightforward.\n[segment 168] Unknown speaker: For example, let's look at this without diving\n[segment 169] Unknown speaker: into technical details.\n[segment 170] Unknown speaker: In the slide, you have a typical Python example\n[segment 171] Unknown speaker: that includes an endpoint where the model lives,\n[segment 172] Unknown speaker: an API key to authenticate securely,\n[segment 173] Unknown speaker: and the prompt submission in the input,\n[segment 174] Unknown speaker: the same prompt you tested.\n[segment 175] Unknown speaker: What is important to know the exact same concept applied\n[segment 176] Unknown speaker: in Python, JavaScript, C#, or any language.\n[segment 177] Unknown speaker: Only the syntax is going to change, not the logic.\n[segment 178] Unknown speaker: And yes; you use OpenAI or APIs through Azure Microsoft Foundry\n[segment 179] Unknown speaker: to consume these models in code.\n[segment 180] Unknown speaker: Up to now, we have been working\n[segment 181] Unknown speaker: with generative AI models testing prompt,\n[segment 182] Unknown speaker: tuning parameters,\n[segment 183] Unknown speaker: and understanding how models generate responses.\n[segment 184] Unknown speaker: The natural next step is turning a model into an agent.\n[segment 185] Unknown speaker: That's exactly what the Microsoft Foundry Agent Service\n[segment 186] Unknown speaker: is for.\n[segment 187] Unknown speaker: What does creating an agent really mean?\n[segment 188] Unknown speaker: In simple terms, a model answers questions; an agent is designed\n[segment 189] Unknown speaker: to get something done.\n[segment 190] Unknown speaker: When you create an agent in Foundry,\n[segment 191] Unknown speaker: you take a model configuration that works;\n[segment 192] Unknown speaker: save it as a named agent; give it persistent instruction; and,\n[segment 193] Unknown speaker: finally, connect it to tools and knowledge\n[segment 194] Unknown speaker: so the agent doesn't forget its role every time you talk to it.\n[segment 195] Unknown speaker: One key difference from models is that agents remember\n[segment 196] Unknown speaker: who they are and what they are supposed to do.\n[segment 197] Unknown speaker: Their instructions act like a job description.\n[segment 198] Unknown speaker: What they help with, what tool to use,\n[segment 199] Unknown speaker: and what they should not do, this makes agent consistent","segments":[{"id":"04b2aa5e-b2f5-4f10-ac90-af963f3fc8ac","segment_index":0,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"[ MUSIC ]"},{"id":"da5799ee-b998-4a21-b689-8e10b83aa950","segment_index":1,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"SARAH ALLALI: Okay."},{"id":"9bdf9a5a-c5c8-46d8-bb90-ef39ebdb3281","segment_index":2,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"I want to build an AI assistant, one that can answer questions,"},{"id":"67fe5a81-0eb9-4ac1-900a-e0e194c4e135","segment_index":3,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"check information, and even take actions."},{"id":"0812b5c7-c058-49e1-b3aa-aaaa83ebfbe3","segment_index":4,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"But now I have a bigger question."},{"id":"c7e61c41-ca4a-4520-acdf-059306006904","segment_index":5,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Which model should I choose?"},{"id":"40455371-1f76-4ede-9d7f-a1a516fefda8","segment_index":6,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"There are so many options, different models,"},{"id":"7d1480e9-4609-4fbf-b831-080c64bfdeca","segment_index":7,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"different capabilities, different providers;"},{"id":"a6bd1f94-7462-468e-b11e-6576630deb20","segment_index":8,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and the choice actually matters"},{"id":"83b2caf2-e7e3-465b-bc02-b5d041f7971d","segment_index":9,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"because the model I choose will directly impact how my agent"},{"id":"cf2adfe5-eb38-4828-b87b-ec7540c3b8d7","segment_index":10,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"going to behave."},{"id":"5c236ae0-0676-4712-93fc-2f2f333c4061","segment_index":11,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"So how do I choose the right model"},{"id":"115bec31-ebb1-4c41-96aa-ac899f2c014d","segment_index":12,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"for the right scenario instead of just guessing?"},{"id":"4c19951b-25ce-4965-94ff-e52212366ee1","segment_index":13,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"What if I had a single place where I can explore, compare,"},{"id":"de13e540-93ce-40c0-804f-955da0252332","segment_index":14,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"test models before using them?"},{"id":"c5ceb7f7-d155-4b72-bb4a-1a0d4004dbcd","segment_index":15,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"That's exactly what Microsoft Foundry comes in."},{"id":"cde1e5e2-c1e6-412d-8311-90813da612eb","segment_index":16,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"If you are wondering how to choose models and use them"},{"id":"ee29b0d6-ad95-449f-961e-1b67dd861c8d","segment_index":17,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"to build a diligent agent, you are in the right place."},{"id":"4f44075b-eae2-4086-8b5f-0a24596e306d","segment_index":18,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Hi. I'm Sarah Allali, a senior technical trainer"},{"id":"85047296-df9f-45fd-b878-4c3be6a16b43","segment_index":19,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"at Microsoft specializing in AI Copilot and building intelligent"},{"id":"ccbc7fe2-56a9-49d8-accc-32a632c2b052","segment_index":20,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and secure application on Microsoft Azure."},{"id":"076ba83c-57e5-42c0-a27f-748031960401","segment_index":21,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"I have a PhD in computer science"},{"id":"e0d43d59-b0b2-4075-8f50-eb4fb125e5c5","segment_index":22,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and 12 years teaching training experience."},{"id":"dca239e3-5aed-4f47-af9e-12df2a56e34a","segment_index":23,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This course covers 12 sessions where we'll dive"},{"id":"4d8f798e-9b75-42c2-acdb-d40d23e7dcc1","segment_index":24,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"into CoreAI concepts and Azure and learn how to turn ideas"},{"id":"2e190557-1ba5-42dc-8695-935063d7aeff","segment_index":25,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"into real applications using modern tools"},{"id":"3b0fda66-d7c1-4dec-9487-9f156b35de61","segment_index":26,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"like Microsoft Foundry."},{"id":"750e97d6-d713-42ef-908d-3042605a094c","segment_index":27,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"In the first session, we will see how to get started"},{"id":"cb341aaf-843e-4b16-872a-22db495b5b97","segment_index":28,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"with generative AI and agents in Azure."},{"id":"8596bbc3-6312-4812-8ab3-835f2cfe495b","segment_index":29,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"I will show you how Microsoft Foundry helps you discover,"},{"id":"e4f73c75-392c-4497-9ef7-7c9335470b3a","segment_index":30,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"evaluate, and use AI models"},{"id":"96e36d33-2dd4-4a34-86e4-ffb5b17d9495","segment_index":31,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"to build real applications and agents."},{"id":"6006d1ad-720b-4ef8-89da-374116c93458","segment_index":32,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Up to now, we have talked about models, agents,"},{"id":"23cb1683-3c5d-4a5d-957f-ace4705f3b1d","segment_index":33,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and how they work together."},{"id":"5dbddd03-7d70-4ddc-bc7a-1871c83610bd","segment_index":34,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Now let's answer a very practical question."},{"id":"641cfbe3-5a23-4763-8321-cbfa242d4f97","segment_index":35,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Where do these models come from,"},{"id":"5ebbf792-c252-4f36-8b18-f71fea339a3c","segment_index":36,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and how do we choose the right one?"},{"id":"1a0bf143-d946-4981-ab3e-dcaeb2449ea9","segment_index":37,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"That's exactly where Microsoft Foundry comes in."},{"id":"e29f3590-9c79-431b-a49d-8058815a3cd1","segment_index":38,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Think of Microsoft Foundry as a created marketplace"},{"id":"7afc8169-d984-4724-ba89-5b4a773f28e6","segment_index":39,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and delivery platform for AI models."},{"id":"6ad5a2bd-8d43-4409-b77d-8cd69281dbeb","segment_index":40,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Instead of randomly picking a model and hoping it works,"},{"id":"d1ffd83b-e86a-43a5-b239-5eea9f9d7d89","segment_index":41,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Foundry gives you a catalog"},{"id":"c6eed272-781c-4b60-8abc-6ba954ed077a","segment_index":42,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"where models are clearly described, categorized,"},{"id":"c466d599-c41b-4ef2-9bb7-09d12ea4b175","segment_index":43,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"governed, and delivered in a safe, enterprise-ready way."},{"id":"4f1c3639-4c1e-4957-92f0-e54acd3cdd65","segment_index":44,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"You are not guessing."},{"id":"3f77a5ae-e224-4de0-9d69-8e7fb8a839fd","segment_index":45,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"You are choosing intentionally."},{"id":"25e6ebe9-7af9-4933-b3cc-700529a99066","segment_index":46,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"At Microsoft Foundry, we provide two big model categories,"},{"id":"3e6fdee1-ab1c-4ab0-a06b-37a247ecf02e","segment_index":47,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"models sold directly by Microsoft and models"},{"id":"40434c1f-2f52-4a5f-968b-52f6c50cdc07","segment_index":48,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"from partner and community."},{"id":"81704912-61c4-4fc6-ab05-4f0b6bcde7c5","segment_index":49,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Models sold directly by Microsoft are models"},{"id":"c6742f27-a654-4df9-b220-f41fc16911ac","segment_index":50,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"that are hosted by Microsoft and built"},{"id":"c0a883fa-02c1-4455-b041-9a7c6ddd7d5a","segment_index":51,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"through your Azure subscription."},{"id":"bd0852e9-542b-42be-9374-299ba08b8d43","segment_index":52,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"They come with enterprise support and SLAs,"},{"id":"e91aa244-5201-4a56-bf01-0641cb450068","segment_index":53,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"follow Microsoft responsible AI standards that we talked"},{"id":"14412575-bba8-4a7f-a377-bb3f68951e65","segment_index":54,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"about in a previous session."},{"id":"eb8253ef-885c-46db-9fb8-6b1d3768fab3","segment_index":55,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"These include all Azure OpenAI models and selected models"},{"id":"5344c619-c4d4-439b-89c6-7c35f5f2184a","segment_index":56,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"from trusted providers that Microsoft manages directly."},{"id":"da1c8934-05fd-4f01-81ac-8c7671e5a449","segment_index":57,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, if you are a company building an internal HR"},{"id":"d7a4e3ce-87ae-484d-869f-6ffe834d5fcf","segment_index":58,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"assistant and you want enterprise security,"},{"id":"30c96465-55ac-4093-8b7e-981ff7d46e79","segment_index":59,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"predictable billing, and Microsoft support,"},{"id":"c4498430-e733-49f5-b8d6-335179ca49c7","segment_index":60,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"you will typically start with Azure OpenAI models in Foundry."},{"id":"3f0d200a-028b-4e2f-a016-3960404f933a","segment_index":61,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The partner community models gives access to models"},{"id":"e454b96b-7606-4f5f-a799-98c973f8823b","segment_index":62,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"from Research Lab, open source communities,"},{"id":"2b959e3e-2f5c-4b51-897d-bc478a1279c4","segment_index":63,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and specialized AI providers."},{"id":"223d0278-f9ba-4b01-b659-aee472a46f34","segment_index":64,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"These models are provided by trusted third-party partners,"},{"id":"a05bf4da-2970-4cf9-93cc-e5d8353a85e2","segment_index":65,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"purchased under the partners' terms, and are useful for niche"},{"id":"20a02202-c5bc-45a9-95df-12853f2d0c66","segment_index":66,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"or specialized use cases."},{"id":"fbcef86f-0d99-4769-adbb-d38ca198a09e","segment_index":67,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, you are building an AI system for medical imaging"},{"id":"e150fb16-aec7-48bd-aa6e-7803a7ff0b5e","segment_index":68,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"or legal document analysis."},{"id":"0a7e453e-bd15-401f-8dcd-7c25dda441ee","segment_index":69,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"A partner or community model may be a better fit"},{"id":"d64d1ed9-6361-41c2-9c6a-239620ed18f8","segment_index":70,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"than a general purpose language model"},{"id":"fd3a88ca-09d9-4967-8c67-ecee532dd9e4","segment_index":71,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"for some specific scenarios."},{"id":"45dc13c4-aa9c-4e20-998c-0e7fd8c5efe8","segment_index":72,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Foundry doesn't just list models."},{"id":"dcfbcb53-df49-4fa3-a493-627e6b3c7336","segment_index":73,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"It helps you compare them intentionally based"},{"id":"94ee431d-7311-4c43-8e06-216229548338","segment_index":74,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"on real decision factors."},{"id":"2d201bc5-ce77-41da-bff9-9379e81cba9f","segment_index":75,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"And let's break those down."},{"id":"6f4af22e-de05-431b-aae3-f97ebed70fa5","segment_index":76,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"First, deployment type meta, where the work happens."},{"id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","segment_index":77,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This answer questions like where is your data processed?"},{"id":"813f22d1-4282-47f0-b0c9-4b5ee51a8c0f","segment_index":78,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"How is the model hosted?"},{"id":"9cf0ed7d-e041-4071-b2e2-a7699b9a3c1f","segment_index":79,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"How do you pay for it?"},{"id":"e3e2fd83-f5b4-4ed9-b52b-5d2f942cf7f4","segment_index":80,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, a public customer-facing chatbot may use"},{"id":"695206a9-794d-472d-be2b-9c5df66b776e","segment_index":81,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"a fully managed cloud deployment."},{"id":"c82de5e5-9e68-4c36-abab-5d9ab2e53069","segment_index":82,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"A sensitive internal tool may require stricter hosting rules."},{"id":"b77444eb-77bf-4875-9196-7a0153ebcb83","segment_index":83,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The second factor is versioning and updates."},{"id":"274c85e0-c5cb-4ce8-87e6-32c4650cbd3e","segment_index":84,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"These depends on your preference."},{"id":"d525308b-f225-4ecf-a979-2501a58740f2","segment_index":85,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Do you prefer stability, or do you prefer innovation?"},{"id":"5a306649-9b98-482b-8a7f-fc46b3577852","segment_index":86,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Models evolve, and Foundry lets you see the model version,"},{"id":"20a1990b-f1a4-4741-92cf-5e2bfec4b6cc","segment_index":87,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"whether it is auto updated or stay fixed for consistency."},{"id":"beab2cf4-a9be-4409-8a42-4b7a4442d807","segment_index":88,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, if you are in a finance or healthcare,"},{"id":"b57821e6-01c6-419d-8fd9-de0140310071","segment_index":89,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"you may prefer a fixed version."},{"id":"d50031a5-8938-4a2e-ad31-67ddc3d6a867","segment_index":90,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"If you are experimenting, you may welcome automatic upgrades."},{"id":"5da1d8c0-1b27-4412-b8dc-5ed79a72b107","segment_index":91,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The third factor is about the rate limits."},{"id":"6bec9834-37ba-4aeb-a5f3-de66019fc429","segment_index":92,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This is all about how fast and how often."},{"id":"69d52b9e-5107-4fc4-92ab-68ce9b25c26c","segment_index":93,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Every model has limits, such as the maximum tokens per minute"},{"id":"6d4e7113-8065-46f7-9fbd-94f48a7bc9a1","segment_index":94,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and the throughput constraints."},{"id":"d852019d-6f12-4e12-be03-879e409c768a","segment_index":95,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"So, for example, a high-volume customer support bots needs"},{"id":"aba2eaa3-13f8-4838-b2e5-2ff675a6d24f","segment_index":96,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"higher rate limits than a small internal assistant."},{"id":"1fcc1dad-8716-4caf-a6a0-696cd0f1cd79","segment_index":97,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The last factor is about guardrails"},{"id":"89928d68-403c-4b2c-b516-94b29e5ea62f","segment_index":98,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and responsible AI policies."},{"id":"4b1bb4b3-3778-4c6c-a373-a150a4aeaa15","segment_index":99,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"And this is critical."},{"id":"94c03f8e-01bf-47db-8736-587bac0e74d7","segment_index":100,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Foundry models come with built-in responsible AI policy,"},{"id":"858ed935-2ad7-407b-8604-ac379051894b","segment_index":101,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"content safety control, and usage guidelines."},{"id":"efa2f1e0-1324-4c6c-baf2-cba1d34f3c81","segment_index":102,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, you don't want your AI assistant generating unsafe"},{"id":"d8013951-7122-46e7-b4a2-3c3f08773586","segment_index":103,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"or inappropriate content."},{"id":"9bbc7aa9-6058-449f-a60c-5ad337108787","segment_index":104,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Foundry helps apply guardrails by design,"},{"id":"47124cda-5a74-4065-88bd-138ba94d34fd","segment_index":105,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"not as an afterthought."},{"id":"ba4f346e-396c-4423-8331-8e479317576c","segment_index":106,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Now that we understand models and agents,"},{"id":"4c32ed4b-98d0-45df-b0d4-ea30dbe4b63b","segment_index":107,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"let's look at how we actually use a generative AI model"},{"id":"d1393b2a-a10f-487c-af06-72da15178a54","segment_index":108,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"in practice."},{"id":"c1cff025-be26-4d17-8aa1-8efde4dde75d","segment_index":109,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"And the most important rule here is simple:"},{"id":"4d62b70e-d3cb-415d-a0f4-8e03d5fdb5d4","segment_index":110,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Before you write code, you experiment."},{"id":"57d446a7-7b06-4cc0-9683-605ceccfaa42","segment_index":111,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"That's exactly why Microsoft Foundry provides the"},{"id":"83d982b7-36ea-4307-b192-09307883c6b4","segment_index":112,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Model Playground."},{"id":"a987da7a-e03e-4a2c-9934-fc5ab9f3e484","segment_index":113,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Think of the Model Playground as a sandbox."},{"id":"21bb53af-9ffc-426f-a88a-2cfc100b7b10","segment_index":114,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"It is a safe place where you can try different models,"},{"id":"14ba75ea-1a1b-4340-a29d-943f132f9dc7","segment_index":115,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"test prompts, adjust settings, and see result instantly,"},{"id":"d0b0da0c-ead3-4450-8602-a2db0a2a8ce8","segment_index":116,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"all without writing a single line of code."},{"id":"8ed2b85a-ea66-4cdf-b51c-a039ce186209","segment_index":117,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, you want an AI"},{"id":"634dcfff-cd47-4080-8722-5c9513d7adef","segment_index":118,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"to help employees write professional emails."},{"id":"e6230e08-dbb6-4356-b226-fd3cc0c8cec1","segment_index":119,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"In the playground, you can type or write a polite email asking"},{"id":"be86b39c-65d5-4f6e-bc14-3286dc061abf","segment_index":120,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"for a deadline extension."},{"id":"61996bf8-2ad3-4564-9e93-a904ed699869","segment_index":121,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"You see the result immediately."},{"id":"d4368b8d-95d5-4604-8a4f-09ff21d6a593","segment_index":122,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Tweak the wording and refine the tone before you build anything."},{"id":"06d10d01-0ecc-42c4-b1e8-c2b9268197ba","segment_index":123,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Why the playground is so valuable?"},{"id":"c5dda91c-bfa0-485a-9b4c-75a22299fbbe","segment_index":124,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Because it helps you to test prompt"},{"id":"7b775d26-593f-4fcc-9e88-6a3cd0857777","segment_index":125,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"by understanding what works and what doesn't work."},{"id":"fbe05089-edbc-4abc-9bf3-f75a50fd8b5c","segment_index":126,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Tune parameter by controlling tone, length, creativity,"},{"id":"ca8981a2-b4fe-4d6d-820d-46c86282b588","segment_index":127,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and also capture configuration"},{"id":"4ffef237-fd24-473c-a1c3-84ec8b86d3f6","segment_index":128,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"by saving exact setting before coding."},{"id":"f0c82117-1693-4dda-97c5-f881cf6324f9","segment_index":129,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This avoids a very common mistake, jumping straight"},{"id":"2de927ae-4dd7-461e-8102-82d0dd004da2","segment_index":130,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"into code and debugging prompt at the same time."},{"id":"2fbff4df-1cb6-46f5-9f03-61e306c95dfb","segment_index":131,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"When you work with a generative AI model,"},{"id":"86dca512-70d7-48d5-9c65-cb0678d9829f","segment_index":132,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"whether in the playground or in code,"},{"id":"c4117811-88c2-4b2f-b80d-4e47eaea8f4a","segment_index":133,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"you always configure three main things: instructions,"},{"id":"1d26fe7e-8672-47b0-9579-34fc08d6f330","segment_index":134,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"input, and parameters."},{"id":"dbfba525-5c0f-4bfc-9f52-697ede4af3cd","segment_index":135,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Instruction, or system prompt, sets the rule"},{"id":"56070b84-4fc3-4d5b-9e12-1a944c5d6c90","segment_index":136,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and the rules for the model."},{"id":"60acb23f-c5e7-43f9-b199-1b604274a963","segment_index":137,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Think of it as giving the AI a job description."},{"id":"94d2bc71-385a-4453-ac9c-9c18fada80b7","segment_index":138,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, for an HR agent, we would tell it,"},{"id":"564f5dc5-0e6d-4821-ba01-5ae5a92b2914","segment_index":139,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"\"You are a helpful HR assistant."},{"id":"9ec24414-f9df-4e89-9a62-44b436560500","segment_index":140,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"You respond professionally and clearly."},{"id":"3b321646-b6ac-4699-9274-8fd034659c2b","segment_index":141,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"You don't give any legal advice."},{"id":"c093053c-3185-4c3d-81da-719650d390d9","segment_index":142,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This frames how the model behaves."},{"id":"2dbc22d0-67ad-4ab7-a1ed-877521b87238","segment_index":143,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Next, we have input, or user prompt."},{"id":"6fa62210-f7fa-4327-b2c9-4928a4ffe7de","segment_index":144,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This is what the user actually asks."},{"id":"62802af4-7266-4e55-9f8d-4cc1f6d3b865","segment_index":145,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Good prompts are always clear, specific, and context-rich."},{"id":"9fd1de94-8c5d-46ba-add0-29898b79eec7","segment_index":146,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, rather than writing, Write an email,"},{"id":"bd7024a8-80fc-41d2-a66a-a0738c0ec16f","segment_index":147,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"in the prompt, you would say,"},{"id":"ca4dd17a-0e63-4ff8-a9b0-c3e4dab173ad","segment_index":148,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Write a friendly email reminding a colleague"},{"id":"c898efe2-5465-4deb-a6b4-e3c3374c20d6","segment_index":149,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"about a meeting tomorrow at 10am."},{"id":"032898b8-875b-4d77-988e-56ad52418848","segment_index":150,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This prompt will get you a much better result."},{"id":"76427f57-f1b2-46e8-9186-fd2f5c5f20e1","segment_index":151,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Finally, the parameters."},{"id":"c6579b3f-a262-400b-b8e9-57ad56dc1f19","segment_index":152,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The parameters is all about how the model responds."},{"id":"caf3f707-42d0-472e-97ee-45aa2b613239","segment_index":153,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Parameters help fine tune the response"},{"id":"76bddb24-6406-4126-8296-a12bcacf63ce","segment_index":154,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"by using factors such as creativity."},{"id":"b076f500-2b3e-401e-9d9c-39e71e9dcfad","segment_index":155,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"If you put it low, you will get you a precise,"},{"id":"f951804f-ef6f-49b2-a462-5821947c84a8","segment_index":156,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"predictable answers."},{"id":"0b379dfc-3a09-4ac3-b0b0-72e2f4f23cd4","segment_index":157,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"And, if you put it high,"},{"id":"c655a7bd-f680-4543-a73a-1512386bd05d","segment_index":158,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"you will produce a more creative, varied response."},{"id":"d15d6d45-0985-435f-b9a4-44141808fdd0","segment_index":159,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Another factor is the response length, like a short"},{"id":"f399712f-f246-4ace-82d7-8f5a85b23e02","segment_index":160,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"or maybe a long or maybe more detailed explanation."},{"id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","segment_index":161,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"So what is the best configuration for the factors?"},{"id":"565bac5b-e2a8-40ea-ba26-351bdcf9bba3","segment_index":162,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"This depends on the context."},{"id":"a70aa1c7-9f67-45c0-bebd-4cc53f3ecc39","segment_index":163,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, for a legal summary,"},{"id":"25bc0ccd-70b6-4d2c-be74-7a8adf97c0c5","segment_index":164,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"low creativity is better; and, for marketing copy,"},{"id":"fbb14ad6-c638-439b-a2d7-ce23d451247d","segment_index":165,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"high creativity works well."},{"id":"f8836bd1-9c0a-4288-a298-7afff16cfe68","segment_index":166,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Once you are happy in the playground,"},{"id":"ea11362c-1f34-4a37-b735-a52b68074ce1","segment_index":167,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"moving to code is straightforward."},{"id":"b707268e-3196-4c54-8f71-e80987e1b993","segment_index":168,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"For example, let's look at this without diving"},{"id":"676f60b1-7eb9-4b6d-9dfd-b915871fb4d5","segment_index":169,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"into technical details."},{"id":"d20b40a1-70e4-4869-9db6-619e80b1d092","segment_index":170,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"In the slide, you have a typical Python example"},{"id":"bf419b9e-2c0d-4727-bf54-b9b646027a6a","segment_index":171,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"that includes an endpoint where the model lives,"},{"id":"dee9acc0-b6b4-4c42-8f2f-730eba99987e","segment_index":172,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"an API key to authenticate securely,"},{"id":"c1ba2853-0d55-4035-a060-ebed807f8884","segment_index":173,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and the prompt submission in the input,"},{"id":"f0eee27a-85f3-4032-8e38-a921e0994ccb","segment_index":174,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"the same prompt you tested."},{"id":"0b466d13-545c-40ae-ad22-6be4c72bc56e","segment_index":175,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"What is important to know the exact same concept applied"},{"id":"b8f1fede-857f-40f3-a015-eccd4062a8f3","segment_index":176,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"in Python, JavaScript, C#, or any language."},{"id":"46574941-733d-4a29-925b-906411701851","segment_index":177,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Only the syntax is going to change, not the logic."},{"id":"5a7692c6-e8f6-4e75-bf13-e0eb0a5b20b8","segment_index":178,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"And yes; you use OpenAI or APIs through Azure Microsoft Foundry"},{"id":"30541cd6-3099-43fa-bc1b-8c250e4fd8e4","segment_index":179,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"to consume these models in code."},{"id":"ec708373-40fb-46c2-ae50-e242b67e60eb","segment_index":180,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Up to now, we have been working"},{"id":"1602f2d6-4e22-47b2-81a8-e346ea6294e0","segment_index":181,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"with generative AI models testing prompt,"},{"id":"8972ce9a-cb00-4326-bcbe-d3e56cab01a1","segment_index":182,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"tuning parameters,"},{"id":"188adcd6-805f-43de-88a5-7b79fa8cbc65","segment_index":183,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and understanding how models generate responses."},{"id":"4da51511-1811-4a4c-8353-47f863967367","segment_index":184,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"The natural next step is turning a model into an agent."},{"id":"70be282e-6d76-4d2b-a57b-3b0d67f26e51","segment_index":185,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"That's exactly what the Microsoft Foundry Agent Service"},{"id":"924884be-174c-4faf-a8ef-822d46805a23","segment_index":186,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"is for."},{"id":"dc6d2a3f-fb84-485d-a734-46b5084f104c","segment_index":187,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"What does creating an agent really mean?"},{"id":"299e374c-63f2-4fcb-b70f-8352272a6e2f","segment_index":188,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"In simple terms, a model answers questions; an agent is designed"},{"id":"6777a085-3ce2-4ff0-b419-594f0bb6af79","segment_index":189,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"to get something done."},{"id":"765f1eb4-daf1-448f-bd92-76b7fd667b84","segment_index":190,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"When you create an agent in Foundry,"},{"id":"6491473a-0c01-42a7-83ad-06e317ac12b6","segment_index":191,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"you take a model configuration that works;"},{"id":"a325c300-3c2d-44be-97fc-7c9ad3aaea47","segment_index":192,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"save it as a named agent; give it persistent instruction; and,"},{"id":"c11f1c25-20a6-4046-bdde-2dc183d710e4","segment_index":193,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"finally, connect it to tools and knowledge"},{"id":"2e4f0a87-0e49-44c2-ad1e-824919da126a","segment_index":194,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"so the agent doesn't forget its role every time you talk to it."},{"id":"9362d173-4c04-4634-b0b9-1525425fb4ee","segment_index":195,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"One key difference from models is that agents remember"},{"id":"1bafd887-1eef-467c-ae79-3ffc345ab7ff","segment_index":196,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"who they are and what they are supposed to do."},{"id":"5d068124-8203-4c8a-b470-b98a003b0801","segment_index":197,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"Their instructions act like a job description."},{"id":"2ebbdfb9-fb96-497d-bfa0-74031ae68ff6","segment_index":198,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"What they help with, what tool to use,"},{"id":"bce4290a-03c3-40d8-a1aa-e8c171ed41f1","segment_index":199,"speaker_name":null,"start_seconds":null,"end_seconds":null,"text":"and what they should not do, this makes agent consistent"}]},"content_assets":[{"id":"17ed7fd4-1f7f-4999-a2eb-c216c31f6876","artifact_kind":"content_asset","artifact_type":"content_calendar_item","title":"Calendar idea 3","status":"ready","generated_at":"2026-07-01T06:07:41.317964Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Calendar idea 3\n\nTurn segment 161 into a follow-up asset focused on the episode signal.","text":"Turn segment 161 into a follow-up asset focused on the episode signal.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"e2349166-235d-4dd0-895b-39a7c9e31320","artifact_kind":"content_asset","artifact_type":"content_calendar_item","title":"Calendar idea 2","status":"ready","generated_at":"2026-07-01T06:07:41.317936Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Calendar idea 2\n\nTurn segment 85 into a follow-up asset focused on the episode signal.","text":"Turn segment 85 into a follow-up asset focused on the episode signal.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"4d78ec68-1b40-4a2b-8f0b-4c90865a4214","artifact_kind":"content_asset","artifact_type":"content_calendar_item","title":"Calendar idea 1","status":"ready","generated_at":"2026-07-01T06:07:41.317906Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Calendar idea 1\n\nTurn segment 77 into a follow-up asset focused on the episode signal.","text":"Turn segment 77 into a follow-up asset focused on the episode signal.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"content_calendar_item","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"d83577b9-f069-4637-a886-10cf892b9c79","artifact_kind":"content_asset","artifact_type":"newsletter_summary","title":"Newsletter summary Get started with generative AI and agents in Azure | AI-901 | Episode 5","status":"ready","generated_at":"2026-07-01T06:07:41.317872Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"newsletter_summary","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Newsletter summary Get started with generative AI and agents in Azure | AI-901 | Episode 5\n\nThis episode reinforced: segment 77, segment 85, segment 161.","text":"This episode reinforced: segment 77, segment 85, segment 161.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"newsletter_summary","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"47778daf-f1d9-4fe3-8000-cd0f17e31aed","artifact_kind":"content_asset","artifact_type":"social_post","title":"Post draft 161","status":"ready","generated_at":"2026-07-01T06:07:41.317835Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Post draft 161\n\nThis episode keeps returning to the market signal.\n\nQuote: \"So what is the best configuration for the factors?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","text":"This episode keeps returning to the market signal.\n\nQuote: \"So what is the best configuration for the factors?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"985fd037-f09f-4403-894e-101ecc43a504","artifact_kind":"content_asset","artifact_type":"social_post","title":"Post draft 85","status":"ready","generated_at":"2026-07-01T06:07:41.317806Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Post draft 85\n\nThis episode keeps returning to the market signal.\n\nQuote: \"Do you prefer stability, or do you prefer innovation?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","text":"This episode keeps returning to the market signal.\n\nQuote: \"Do you prefer stability, or do you prefer innovation?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"b91f7705-24e6-4d3e-a231-c1272f57c826","artifact_kind":"content_asset","artifact_type":"social_post","title":"Post draft 77","status":"ready","generated_at":"2026-07-01T06:07:41.317776Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Post draft 77\n\nThis episode keeps returning to the market signal.\n\nQuote: \"This answer questions like where is your data processed?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","text":"This episode keeps returning to the market signal.\n\nQuote: \"This answer questions like where is your data processed?\"\n\nWhy it matters now: it connects directly to this week's ranked themes.","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"post_draft","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"27cedf26-ef19-401a-a38f-6fb1b905cfff","artifact_kind":"content_asset","artifact_type":"hook","title":"Reel hook 161","status":"ready","generated_at":"2026-07-01T06:07:41.317746Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Reel hook 161\n\nLead with this line: \"So what is the best configuration for the factors?\"","text":"Lead with this line: \"So what is the best configuration for the factors?\"","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"792e1734-9b68-48ec-bc8e-5a8bb87c0a1c","artifact_kind":"content_asset","artifact_type":"hook","title":"Reel hook 85","status":"ready","generated_at":"2026-07-01T06:07:41.317718Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Reel hook 85\n\nLead with this line: \"Do you prefer stability, or do you prefer innovation?\"","text":"Lead with this line: \"Do you prefer stability, or do you prefer innovation?\"","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"ee5aa549-46b4-4136-bdf5-76173767fbca","artifact_kind":"content_asset","artifact_type":"hook","title":"Reel hook 77","status":"ready","generated_at":"2026-07-01T06:07:41.317689Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Reel hook 77\n\nLead with this line: \"This answer questions like where is your data processed?\"","text":"Lead with this line: \"This answer questions like where is your data processed?\"","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"reel_hook","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"5b38591e-8574-4e42-9723-73a553973af8","artifact_kind":"content_asset","artifact_type":"quote_card","title":"Quote moment 161","status":"ready","generated_at":"2026-07-01T06:07:41.317657Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Quote moment 161\n\nSo what is the best configuration for the factors?","text":"So what is the best configuration for the factors?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"68d69349-0804-4537-b545-8056357ab82f","artifact_kind":"content_asset","artifact_type":"quote_card","title":"Quote moment 133","status":"ready","generated_at":"2026-07-01T06:07:41.317628Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"c4117811-88c2-4b2f-b80d-4e47eaea8f4a","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.24,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Quote moment 133\n\nyou always configure three main things: instructions,","text":"you always configure three main things: instructions,","structured":{"metadata":{"reasons":[],"clip_score":0.24,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"0f2d8680-fcc8-4d7e-987c-81525e5ffef3","artifact_kind":"content_asset","artifact_type":"quote_card","title":"Quote moment 85","status":"ready","generated_at":"2026-07-01T06:07:41.317599Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Quote moment 85\n\nDo you prefer stability, or do you prefer innovation?","text":"Do you prefer stability, or do you prefer innovation?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"54355c9b-60af-425a-a31b-13f8528e1559","artifact_kind":"content_asset","artifact_type":"quote_card","title":"Quote moment 77","status":"ready","generated_at":"2026-07-01T06:07:41.317569Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Quote moment 77\n\nThis answer questions like where is your data processed?","text":"This answer questions like where is your data processed?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"quote_moment","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"794c5c0a-4b45-44ba-91f3-cde1d20f4aa2","artifact_kind":"content_asset","artifact_type":"clip_candidate","title":"Clip candidate 133","status":"ready","generated_at":"2026-07-01T06:07:41.317535Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"c4117811-88c2-4b2f-b80d-4e47eaea8f4a","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.24,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Clip candidate 133\n\nyou always configure three main things: instructions,","text":"you always configure three main things: instructions,","structured":{"metadata":{"reasons":[],"clip_score":0.24,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"eb03e2cd-ad2b-41d9-abf3-f3e4b1d79ca4","artifact_kind":"content_asset","artifact_type":"clip_candidate","title":"Clip candidate 161","status":"ready","generated_at":"2026-07-01T06:07:41.317506Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Clip candidate 161\n\nSo what is the best configuration for the factors?","text":"So what is the best configuration for the factors?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"5a56a28c-00ca-4c71-9843-df6bb96d12aa","artifact_kind":"content_asset","artifact_type":"clip_candidate","title":"Clip candidate 85","status":"ready","generated_at":"2026-07-01T06:07:41.317477Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Clip candidate 85\n\nDo you prefer stability, or do you prefer innovation?","text":"Do you prefer stability, or do you prefer innovation?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"967c16f4-5cd7-4c90-910e-37d4f4527600","artifact_kind":"content_asset","artifact_type":"clip_candidate","title":"Clip candidate 77","status":"ready","generated_at":"2026-07-01T06:07:41.317445Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Clip candidate 77\n\nThis answer questions like where is your data processed?","text":"This answer questions like where is your data processed?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"clip_candidate","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"0fc56167-aec5-4e73-bd5d-6688d451d39a","artifact_kind":"content_asset","artifact_type":"episode_theme","title":"So What Is The Best Configuration","status":"ready","generated_at":"2026-07-01T06:07:41.317409Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"f977f8eb-cee8-4e9c-a64e-565638eb9249","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# So What Is The Best Configuration\n\nTranscript-backed episode theme candidate: So what is the best configuration for the factors?","text":"Transcript-backed episode theme candidate: So what is the best configuration for the factors?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"e77810c2-2092-49ac-9828-1c24255d0224","artifact_kind":"content_asset","artifact_type":"episode_theme","title":"Do You Prefer Stability, Or Do","status":"ready","generated_at":"2026-07-01T06:07:41.317367Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"d525308b-f225-4ecf-a979-2501a58740f2","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# Do You Prefer Stability, Or Do\n\nTranscript-backed episode theme candidate: Do you prefer stability, or do you prefer innovation?","text":"Transcript-backed episode theme candidate: Do you prefer stability, or do you prefer innovation?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}},{"id":"7993176f-4dbe-40af-b660-f8dc3aa1e75a","artifact_kind":"content_asset","artifact_type":"episode_theme","title":"This Answer Questions Like Where Is","status":"ready","generated_at":"2026-07-01T06:07:41.317252Z","model_provider":null,"model_name":null,"memo_id":null,"episode_id":"fe4a2424-e193-43a6-b472-dadb11b69431","transcript_segment_id":"fd6ad77e-2cd7-4890-a65b-f6a13c5a26e1","ranked_theme_id":null,"theme_snapshot_id":null,"ranked_theme_ids":[],"details_json":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}},"formats":{"markdown":"# This Answer Questions Like Where Is\n\nTranscript-backed episode theme candidate: This answer questions like where is your data processed?","text":"Transcript-backed episode theme candidate: This answer questions like where is your data processed?","structured":{"metadata":{"reasons":[],"clip_score":0.26,"input_hash":"sha256:9fb1e32c25d284b40e93d46bb8534666697eb9e6f3f3ccbb1e394c07ba0920c1","quote_score":0.2,"asset_subtype":"episode_theme","pipeline_version":"prompt12_v1","matched_ranked_theme_id":null,"matched_signal_categories":[],"matched_theme_snapshot_id":null}}}}]}