RAG Deep Dive
Duration: 2 Days
Description
Retrieval-augmented generation (RAG) grounds language models in your own data, and doing it well is one of the most valuable skills in applied AI. This in-depth course goes well beyond a basic RAG pipeline to cover the techniques that make retrieval accurate and reliable: chunking and embedding strategies, advanced retrieval and reranking, evaluation, and handling the failure modes that plague naive implementations. Through hands-on exercises, participants build and systematically improve a RAG system and learn to diagnose why retrieval succeeds or fails.
Class time is divided between instructor-led explanation (35 percent) and hands-on development (65 percent).
Audience
Developers and data/ML practitioners building systems that ground LLMs in custom data. Best suited to those with prior AI development or ML experience.
Objectives
- Explain how retrieval-augmented generation works and why it matters
- Design effective chunking and embedding strategies
- Implement and tune retrieval and reranking
- Evaluate RAG quality systematically
- Diagnose and fix common RAG failure modes
- Build an accurate, reliable RAG system
Prerequisites
Participants should have prior AI development or ML experience. The following will help:
- Working knowledge of Python (see Python Foundations)
- Experience building LLM applications (see Building LLM Applications) or ML foundations
- Familiarity with embeddings and vector search is helpful
Course Outline
- Module 1 – Course Introduction
- Welcome and objectives
- Why RAG: grounding and its limits
- Anatomy of a RAG system
- Module 2 – Preparing Knowledge
- Document processing and chunking strategies
- Embeddings: choosing and using models
- Metadata and structure
- Hands-on exercise: build a knowledge base
- Module 3 – Retrieval
- Vector search fundamentals
- Hybrid and keyword retrieval
- Reranking and filtering
- Hands-on exercise: improve retrieval quality
- Module 4 – Generation and Grounding
- Prompting with retrieved context
- Citations and faithfulness
- Handling missing or conflicting information
- Reducing hallucination
- Module 5 – Evaluation and Failure Modes
- Measuring retrieval and answer quality
- Building evaluation sets
- Common failure modes and fixes
- Hands-on exercise: evaluate and tune a RAG system
- Module 6 – Course Wrap-Up
- Key takeaways
- Advanced directions
- Resources for continued learning