A hands-on workshop on turning retrieval-augmented generation into agentic AI flows - Alexey Grigorev

​Description: ​Retrieval-Augmented Generation (RAG) is a common approach for building AI assistants, but most implementations stop at a single search + answer pattern. In this practical workshop, we go a step further.

​You'll learn how to make RAG pipelines more agentic: enabling decision-making, tool use, multi-step reasoning, and real-time interaction with users. We’ll build step by step, from simple retrieval to agent-powered assistants capable of asking follow-up questions, running tool calls, and updating their own knowledge base.

​What we’ll cover:

• ​Build a basic RAG pipeline: Implement search + prompt + LLM logic using course FAQs
• ​Add agentic behavior - ​Let the assistant decide when to search or answer - ​Introduce state\, reasoning\, and multi-step flows
• ​Implement agentic search - Iterate over multiple search queries - Use search history and reasoning to refine results
• ​Use OpenAI function calling - Add tool support: search and FAQ updates - Maintain interaction history and responses
• ​Build a working assistant - Organize your logic with chat_assistant.py - Interact with a live\, modular AI assistant
• ​Explore PydanticAI - Define tools with type hints and docstrings - Run your assistant in a fully typed agent framework

​What you’ll get: ​A working RAG-based AI assistant that can:

• ​Search FAQs
• ​Decide when to act
• ​Ask follow-up questions
• ​Add entries to its own knowledge base

​Everything runs in Jupyter with Python. You can follow along with GitHub Codespaces or use your own environment. ​ About the Speaker: ​Alexey Grigorev is the creator of DataTalks.Club and instructor of the popular Zoomcamp series. He designs and teaches practical AI/ML workflows used by thousands of learners globally. In this workshop, Alexey brings his hands-on teaching style to the emerging field of agentic LLMs, helping you build useful assistants from scratch.

​ Join our slack: https://datatalks.club/slack.html