Skip to content

09 | Learning Resources

Azure AI Studio is currently In Preview

Preview versions are provided without a service level agreement, and they are not recommended for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews.

This page provides an evolving set of links to documentation, training, samples and other resources that may be helpful to self-guided learners on this journey. Check back often for updates since the platform capabilities and tools are evolving fast.

Azure AI Documentation

  1. Azure AI Services - Pre-built AI services available for deployments (API key, endpoints)
  2. Azure AI Resource - Top-level resource to admin billing, access, policies, resource usage
  3. Azure AI Project - Resource to organize work & save state when building AI solutions
  4. Azure AI SDK - Generative AI SDK, drive code-based interactions with Azure AI platform
  5. Azure AI CLI - .NET implementation, drive command-line interactions with Azure AI platform
  6. Azure AI Studio - UI, drive web-based interactions with the Azure AI platform.
  7. Using vector search in Azure Cognitive Search - Tutorial.
  8. Deploy a web app for chat on your data - Tutorial.
  9. Moderate text and images with content safety in Azure AI Studio - Quickstart

Azure AI Samples

Fork and explore these samples to get hands-on experience with the SDK and CLI in different contexts. Star the repo if you find it useful!

  1. Build a Copilot with Your Data - with Python SDK
  2. Build a Copilot with Your Data - using Promptflow
  3. Build a Copilot with Your Data - using LangChain
  4. Build a Copilot with Your Data - using Semantic Kernel
  5. Build a Production RAG - using PromptFlow, with LLM Ops

OpenAI Cookbook

The OpenAI Cookbook has example code and guides for accomplishing common tasks with the OpenAI API. Most code examples are written in Python but applicable to others.

This collection includes the following Azure-centric samples:

  1. Azure OpenAI chat completions example - for: multi-turn, content filtering
  2. Azure OpenAI completions example - for: single prompt, content filtering
  3. Azure OpenAI embeddings example - for: converting nlp input to numerical representation.
  4. Azure OpenAI functions example - for: function definition, function calling
  5. Azure OpenAI chat completion models with your own data (preview) - for: RAG pattern
  6. Azure AI Search as vector database for OpenAI embeddings - for: similarity, hybrid, semantic

The main concepts covered are:

  • Completions - 'question-answer' text generation with {model, prompt} as inputs
  • Chat Completion - 'multi-turn' text generation with {model, messages} as inputs
  • Embeddings - vector representation, meassures 'relatedness' (similarity) of text
  • Functions - used to connect LLMs to external tools (convert NLP inputs into API calls)

Messages: are defined by {role, content} where the role can be system (context), assistant (response) or user (question). Using the Chat Completions API is recommended (works with latest models) where a "completion" is effectively a single message containing a user question.

Function Calling: where the function is a "tool" (code) that takes a string input (JSON), executes the call (to API), and returns the response as a string (JSON). Chat completion request has a "tools" argument to pass in relevant functions. The model can choose to call one or more functions which return response as a new "message" that model can use for summarizing results in its response.

Embeddings: can be used for: search (ranking), clustering (topics), recommendations (items), anomaly detection )(outliers), diversity measurement (similairity), and classification (labels).

Next Up: Summary & Wrap-Ups