RAG, AI Agents & MCP

An advanced AI course: RAG with embeddings and vector databases, AI agents, the Model Context Protocol (MCP), evaluation, privacy and production. Each lesson has a task and a downloadable solution.

Advanced · 13 lessons

Retrieval-Augmented Generation (RAG)

  1. What Is RAG (Retrieval-Augmented Generation)? - Learn why models need your own data, how RAG works from question to answer, and build a help-center bot that answers only from your documents.
  2. Chunking, Embeddings and Vector Search - Cut documents into overlapping chunks, understand embeddings and cosine similarity, and build a vector search page where you can change chunk size and see the results change.
  3. Vector Databases: pgvector, Qdrant, Pinecone and Chroma - Learn what a vector database really does, how approximate search works, how to choose between the main options, and build a vector store with upsert, delete and filters behind a clean interface.
  4. Advanced RAG: Hybrid Search, Reranking and Citations - Combine keyword and vector search, fuse the rankings, rerank the best results, drop weak matches and make the model cite sources that the server checks.

AI agents

  1. AI Agents Explained: Loops, Tools, Memory and Limits - Understand the difference between a workflow and an agent, build the agent loop with a step limit and a token budget, and give the agent memory that lasts between conversations.
  2. Multi-Agent Systems, Frameworks and the Claude Agent SDK - Learn when several specialised agents help, build a supervisor, researchers, writer and reviewer team with hand-offs and a revision limit, and see how frameworks fit around the same ideas.

Model Context Protocol (MCP)

  1. What Is MCP (Model Context Protocol)? - Learn what MCP is, how hosts, clients and servers fit together, what tools, resources and prompts are, and build an MCP client in Node that starts a server and calls its tool.
  2. Build an MCP Server in Node.js and Connect It to Claude - Build a resume-tools MCP server with two tools and a resource, handle tool errors the right way, and connect it to Claude so the model can call your tools through MCP.
  3. MCP Security: Permissions, Consent and Prompt Injection - Learn the real risks of connecting AI to tools, and build a policy layer with allow, approve and deny rules, a human approval step and a defence against prompt injection from tool results.

Evaluation, safety and production

  1. Evaluating LLM Applications: Test Sets, Metrics and AI Judges - Learn to test an AI app like any other software: build a golden question set, measure retrieval and answers, use an AI judge carefully, and find and fix a real failure.
  2. AI Safety, Privacy and Responsible Use - Learn how to reduce hallucinations, protect personal data by redacting it before it reaches an AI provider, and think about fairness, consent and the law when AI touches people's careers.
  3. Running AI in Production: Cost, Reliability and Monitoring - Make an AI feature fast, affordable and dependable: cache repeated questions, rate limit users, set timeouts, fall back gracefully, and measure tokens, speed and cost.
  4. Your AI Learning Roadmap and Project Ideas - Review what you have built, plan your next steps, choose a portfolio project and write a project plan with data, design, safety, evaluation and cost before you write code.