A practical, production-oriented guide to retrieval-augmented generation — from why AI models fail with live data to the decisions that make RAG systems actually work.
The Series
Part 1: Why AI Gets Things Wrong
Frozen knowledge, no live system access, and why fine-tuning doesn't fix the knowledge currency problem.
Part 2: What RAG Is and Why It Works
RAG as a pattern — retrieve first, then generate. The six components and the line between knowledge and reasoning.
Part 3: How RAG Works — The Complete Pipeline
The full RAG pipeline step by step — ingestion, chunking, embedding, retrieval, augmentation, and generation.
Part 4: Chunking, Retrieval, and the Decisions That Break RAG
Chunking, retrieval, and reranking — the decisions that separate demos from production systems.
Part 5: Build a RAG System in Practice
What happens when a simple RAG pipeline meets real documents — four document shapes, four failure modes, and the decisions each one teaches.
Part 6: RAG, Fine-Tuning, or Long Context?
When to reach for RAG, when to fine-tune, when to lean on long context — and when to combine them.
Part 7: Your RAG System Is Wrong. Here's How to Find Out Why.
Evaluation, faithfulness, and the diagnostic discipline that separates working RAG from broken RAG.
Part 8: RAG in Production — What Breaks After Launch (in drafting — publishing late April)
The production article: data freshness, embedding drift, security, caching, and what actually breaks once your RAG system goes live.
This series is actively maintained. New parts will be linked here as they publish.
Top comments (0)