Part 1 of a 4-part series on how search actually works inside RAG: the retriever architecture, keyword vs semantic search, and why metadata filtering is the only true hard filter.

Part 2 of a 4-part series on information retrieval: how keyword search turns documents into sparse vectors, builds an inverted index, and ranks results with TF-IDF and BM25.

Part 3 of a 4-part series on information retrieval: how semantic search matches documents by meaning instead of words, what embeddings are, how distance is measured, and how contrastive training teaches a model where to place things.

Part 4 of a 4-part series: how to merge keyword and semantic results with Reciprocal Rank Fusion, tune the balance with K and Beta, and measure retrieval quality with Precision, Recall, MAP and MRR.

Part 1 of a 3-part series on Retrieval-Augmented Generation: why LLMs hallucinate on private or fresh information, and how the two-step Retrieval + Generation loop fixes it.

Part 2 of the RAG series: next-token prediction, tokens, autoregression, training on trillions of words — and why that design makes hallucination inevitable without retrieval.

Final part of the RAG series: how a retriever works — knowledge bases, indexes, similarity ranking, the recall/precision tradeoff, and why vector databases matter at scale.

Tigerless Labs' autoharness lets Claude Code learn skills from your real sessions, merge near-duplicates, and archive what stops being used — daemon-free and without a benchmark. Here's how it works and why it matters.