My colleague and I just wrapped up a live series on Python + AI, a nine-part journey diving deep into how to use generative AI models from Python. I gave the english streams while my colleague Gwen gave the spanish streams (and I hung out in her live chat, working on my technical spanish!).
The series introduced multiple types of models, including LLMs, embedding models, and vision models. We dug into popular techniques like RAG, tool calling, and structured outputs. We assessed AI quality and safety using automated evaluations and red-teaming. Finally, we developed AI agents using popular Python agents frameworks and explored the new Model Context Protocol (MCP).
To apply the concepts, we put togethe…
My colleague and I just wrapped up a live series on Python + AI, a nine-part journey diving deep into how to use generative AI models from Python. I gave the english streams while my colleague Gwen gave the spanish streams (and I hung out in her live chat, working on my technical spanish!).
The series introduced multiple types of models, including LLMs, embedding models, and vision models. We dug into popular techniques like RAG, tool calling, and structured outputs. We assessed AI quality and safety using automated evaluations and red-teaming. Finally, we developed AI agents using popular Python agents frameworks and explored the new Model Context Protocol (MCP).
To apply the concepts, we put together code examples that run for free thanks to GitHub Models, a service that provides free models to every GitHub account holder for experimentation and education. The examples are also compatible with local models (via Ollama), Azure OpenAI, or OpenAI.com models.
Even if you missed the live series, you can still access all the material using the links below! If you’re an instructor, feel free to use the slides and code examples in your own classes.
In this session, we explore Large Language Models (LLMs), the models that power ChatGPT and GitHub Copilot. We use Python to interact with LLMs using popular packages like the OpenAI SDK and LangChain. We experiment with prompt engineering and few-shot examples to improve outputs. We also demonstrate how to build a full-stack app powered by LLMs and explain the importance of concurrency and streaming for user-facing AI apps.
In our second session, we dive into a different type of model: the vector embedding model. A vector embedding is a way to encode text or images as an array of floating-point numbers. Vector embeddings enable similarity search across many types of content. In this session, we explore different vector embedding models, such as the OpenAI text-embedding-3 series, through both visualizations and Python code. We compare distance metrics, use quantization to reduce vector size, and experiment with multimodal embedding models.
In our third session, we explore one of the most popular techniques used with LLMs: Retrieval Augmented Generation. RAG is an approach that provides context to the LLM, enabling it to deliver well-grounded answers for a particular domain. The RAG approach works with many types of data sources, including CSVs, webpages, documents, and databases. In this session, we walk through RAG flows in Python, starting with a simple flow and culminating in a full-stack RAG application based on Azure AI Search.
Our fourth session is all about vision models! Vision models are LLMs that can accept both text and images, such as GPT-4o and GPT-4o mini. You can use these models for image captioning, data extraction, question answering, classification, and more! We use Python to send images to vision models, build a basic chat-with-images app, and create a multimodal search engine.
In our fifth session, we discover how to get LLMs to output structured responses that adhere to a schema. In Python, all you need to do is define a Pydantic BaseModel to get validated output that perfectly meets your needs. We focus on the structured outputs mode available in OpenAI models, but you can use similar techniques with other model providers. Our examples demonstrate the many ways you can use structured responses, such as entity extraction, classification, and agentic workflows.
This session covers a crucial topic: how to use AI safely and how to evaluate the quality of AI outputs. There are multiple mitigation layers when working with LLMs: the model itself, a safety system on top, the prompting and context, and the application user experience. We focus on Azure tools that make it easier to deploy safe AI systems into production. We demonstrate how to configure the Azure AI Content Safety system when working with Azure AI models and how to handle errors in Python code. Then we use the Azure AI Evaluation SDK to evaluate the safety and quality of output from your LLM.
In the final part of the series, we focus on the technologies needed to build AI agents, starting with the foundation: tool calling (also known as function calling). We define tool call specifications using both JSON schema and Python function definitions, then send these definitions to the LLM. We demonstrate how to properly handle tool call responses from LLMs, enable parallel tool calling, and iterate over multiple tool calls. Understanding tool calling is absolutely essential before diving into agents, so don’t skip over this foundational session.
In the penultimate session, we build AI agents! We use Python AI agent frameworks such as the new agent-framework from Microsoft and the popular LangGraph framework. Our agents start simple and then increase in complexity, demonstrating different architectures such as multiple tools, supervisor patterns, graphs, and human-in-the-loop workflows.
In the final session, we dive into the hottest technology of 2025: MCP (Model Context Protocol). This open protocol makes it easy to extend AI agents and chatbots with custom functionality, making them more powerful and flexible. We demonstrate how to use the Python FastMCP SDK to build an MCP server running locally and consume that server from chatbots like GitHub Copilot. Then we build our own MCP client to consume the server. Finally, we discover how easy it is to connect AI agent frameworks like LangGraph and Microsoft agent-framework to MCP servers. With great power comes great responsibility, so we briefly discuss the security risks that come with MCP, both as a user and as a developer.