Azure AI Foundry (AI Hub) : Prompt Flow, LLMOps & RAG

Explore generative AI and agents, and understand foundation models and large language models. See how compound AI systems orchestrate tools, plugins, and enterprise knowledge through dynamic planning and human-in-the-loop approaches.

Explore how Azure AI Foundry Studio Hub unifies governance, collaboration, LM Ops, and DevOps to build enterprise AI agents from scratch, fitting into the Microsoft 365 Copilot ecosystem.

Compare hub-based and standalone Microsoft Foundry projects, detailing SDK version differences, collaboration scope from organization to team level, and shared versus proprietary connections to Azure services.

Explore hub-based and standalone Azure AI Foundry projects, compare the classic and new portals, and connect code via SDK or project string to access models, agents, and prompt flow.

Explore how Azure OpenAI provides a tenant-isolated, secure cloud platform that quickly onboards OpenAI models, offering private endpoints, keys, and the Azure OpenAI Studio for deployment.

Explore generative AI jargons like tokens, system and user prompts, and chat completions; learn how unimodal and multimodal models affect input, output, and costs.

Craft precise, detailed prompts to guide generative AI like Copilot and ChatGPT, incorporating the four parts—goal, context, expectations, and source—for accurate, focused results.

Deploy a hub-based Foundry project inside an Azure AI hub to enable prompt flow, and organize with a parent hub and nested projects and a resource group.

Deploy chat completion models from the Azure AI Foundry model catalog and test them in the chat playground, using global standard deployments with token-based pricing.

Discover the Azure AI Foundry architecture, featuring the AI hub with multiple projects, the management center, and secure connections to storage, Azure OpenAI, and large language models via managed identities.

Explore Microsoft prompt flow, a low-code, python-runtime microservice that streamlines building, prototyping, testing, and deploying AI applications powered by large language models. Visualize execution with a right-hand graph, debug and collaborate with versioning and variants, and evaluate pre- and post-deployment performance, including relevance, accuracy, and grounding, all within Azure AI Foundry.

Explore how Microsoft Prompt Flow fits into the Copilot stack, linking foundational models, data, and AI infrastructure with the UI/UX to enable enterprise chatbots and multi-cloud orchestration.

Conduct a hands-on lab to build a named entity recognition prompt flow with an llm component and python data-cleansing. Deploy GPT 4.1, define inputs, run the flow, and view outputs.

Deploy the named entity recognition promflow to a real-time managed endpoint, securing it with primary and secondary keys, and validate it via in-browser testing and REST API calls.

Query PromFlow's real-time endpoint via rest calls, sending a JSON body with entity-type and user-query; compare Paris and France accuracy to Azure Language Service.

Explore a real-life use case of Microsoft Prompt Flow to power an investment recommendation app with persona-based prompts, real-time data, and consumable APIs.

Explore retrieval augmented generation (RAG) to build an enterprise chatbot by combining a retrieval pipeline, grounding knowledge, and an augmented prompt, powered by vector embeddings and Azure AI services.

Explore multimodal retrieval augmented generation with Azure AI Search, combining text, images, and structured documents, preserving layout and citations through image verbalization and layout detection.

Deploy a multimodal rag pipeline on Azure by provisioning storage, enabling anonymous access, uploading Maggie's Travel documents, and configuring an AI search resource with GPT-4 and Ada002 for prompt flow.

Create a multi-modal rag ai search index by connecting Azure Blob Storage, ingesting text and images, and vectorizing with ai foundry text embeddings, image verbalization, and semantic re-ranker.

Create a rag prompt flow in Azure ai foundry by connecting the ai search index, configuring the index lookup with semantic search and Azure OpenAI embeddings, then generate responses.

Discover how function calling enables GPT models to fetch real-time data by calling external APIs within prompt flows, and define functions with a name, description, and required parameters.

Demonstrate integrating function calling in Microsoft Prompt Flow to fetch air pollutant concentrations via a weather API, convert results to a summary, and determine London's air quality index.

Test the weather API in Postman by calling current.json at api.weatherapi.com/v1 with API key and location, enabling AQI, then prepare air quality data for a prompt flow using function calling.

Build a function calling prompt flow in Azure AI Studio with a GPT-4 model, define functions, connect Python code to fetch air quality data, and summarize results.

Build multimodal generative AI applications by blending predictive AI and generative AI, using Azure OpenAI resources, computer vision, and Document Intelligence to create chatbots and analytics workflows.

Build a multi-modal image analysis prompt flow using the computer vision resource and image analysis API to generate dense captions and guide the GPT engine to answer user queries.

Launch and test a computer vision resource with Vision Studio, connect it, and call the api endpoint using the primary key to generate dense captions and detect objects.

Build a multimodal image analysis prompt flow in Azure AI Studio that inputs a user query and image URL, analyzes the image with Python, and returns an LLM response.

Learn to build a prompt flow that lets a GPT-4 model fetch real-time web content via a serp scraping API. Use external data so responses extend beyond training data.

Explore a PII protection prompt flow for healthcare using Azure language service to redact personally identifiable information in LLM outputs with redacted text via named entity recognition.

Provision and test the Azure AI language service using Language Studio, exploring PII redaction and Postman API calls. Learn to configure endpoints and keys, and analyze PII entities in text.

Build a PII protection prompt flow in Azure AI Studio by passing a user query through an LLM, then redact PII with a Python component using Azure Language Service.

Create a chat bot on custom documents using Azure Document Intelligence with prebuilt or custom models. Use GPT-4 in Azure OpenAI to summarize extracted data and answer user queries.

Explore Azure's Document Intelligence resource and its Document Intelligence Studio, test prebuilt invoice models, and learn to extract key-value data from documents using API calls and custom models.

Build a document-aware chat bot in Azure AI Studio using prompt flow to analyze invoices and respond to user queries.

Identify evaluation approaches in azure ai studio: automated, assisted, and manual. Prepare ground truth and expected answers in a dataset, and measure precision, recall, and f1 with llm evaluation.

Run automated evaluation on a rag prompt flow using a travel dataset to assess groundedness with a Likert-scale evaluator, mapping context, query, and answer in Azure AI Foundry.

In this hands-on lab, evaluate and compare models for a RAC pipeline using automated evaluation, a dataset-backed prompt flow, and groundedness criteria to select the best fit model.

Conduct a hands-on custom evaluation on a prompt flow, using a groundedness metric with a question–context–answer dataset from Excel, and review outputs and scores in the matrix.

Explore the fundamentals of responsible AI, including fairness, privacy and security, inclusiveness, transparency, and accountability, plus a four-stage process to identify, measure, mitigate harms, and operate AI workloads.

Explore Azure Content Safety Studio for text, image, and multimodal moderation with severity thresholds. Implement groundedness detection, prompt shields, and protected material detection to safeguard chatbots.

Learn to call the Azure Content Safety Studio API to analyze text and images via Postman. Apply threshold-based guardrails for self-harm, hate, and violence before sending queries to chat engine.

Integrate azure content safety studio with the microsoft prompt flow to filter prompts for self-harm, hate, violence, and adult content in the RAG prompt. Explore configuring a project-level content safety connection, text analysis, a take-action python tool, and conditional activation of index lookup and augmentation in the chat workflow.