Applied AI

Reason + Act: the Thought → Action → Observation loop behind every modern AI agent. From the 2022 research to production practice — with an interactive trace you can step through, move by move.

Learn Model Context Protocol from scratch — how it works, how to build servers in Python, and how to give AI assistants safe access to your files, databases, and APIs. Ends with a complete hands-on project.

Move beyond keyword matching — dense vector search, embedding models, hybrid retrieval, and RAG pipelines that actually ground LLM responses in real knowledge.

Start from 'what is RAG and why' and go all the way to a retrieval system you can trust in production — ingestion, chunking, retrieval quality, grounding, evaluation, and the hard edge cases (stale data, access control, prompt injection, cost). With interactive playgrounds to make it click.

Build AI agents from scratch — single-agent patterns, multi-agent pipelines, memory, observability with Langfuse, and a complete hands-on project. Covers LangGraph, CrewAI, AutoGen, and Agno with real code examples.

Give stateful, human-in-the-loop LangGraph agents a durable memory on stateless AWS Lambda by building a custom DynamoDB checkpointer. From the mental model to a production-ready, multi-tenant implementation you can deploy.

A practical tour of what 'AI memory' actually means, how the tools people reach for day to day — claude-mem, mem0, Letta, LangGraph checkpointers, and the memory built into ChatGPT and Claude — really work, and when to choose which. Architect-level, hands-on, and honest about what's still unsolved.

The capstone decision framework: given a real problem, what do you actually reach for — a raw model call, RAG, tools, MCP, an agent loop, memory, or a Skill — and why, with honest cost trade-offs at both enterprise and bootstrapped scale. Architect-level, hands-on, and built to stop you overbuilding.