RAG & Retrieval
Getting the right few thousand tokens in front of the model, which is most of what makes a knowledge product work.
7 topics
RAG in One Picture
Index your documents once, retrieve the few relevant pieces at question time, put them in the prompt — and accept that retrieval quality, not the model, decides whether it works.
core25 minChunking Strategies
How you split documents decides what can be retrieved at all — and the fixed-size split everyone starts with is the most common cause of bad answers.
core20 minVector Stores & Indexing
Approximate nearest-neighbour search over millions of vectors, and why "just add pgvector to Postgres" is the right first answer more often than a dedicated database.
core20 minHybrid Search & Reranking
Vector search alone misses exact terms and keyword search alone misses meaning — run both, fuse the results, then rerank the survivors with a model that reads them properly.
core20 minEvaluating Retrieval Quality
Measure the retriever separately from the generator, because a bad answer is usually a retrieval miss and you cannot tell by reading the answer.
core20 minCitations & Attribution
Linking each claim back to the span that supports it — the feature that makes a knowledge product trustworthy and, not coincidentally, more accurate.
core15 minWhen You Don't Need RAG
Long context, agentic search, and a plain database query have each eaten a chunk of what RAG used to be the only answer for.
core15 min