hallucination mitigation

RAG is the safer, cheaper path to grounded LLM answers, and this Linux-first guide shows how to run it locally today: install a private stack (fastembed + FAISS + Ollama), build/query indexes with Python scripts, and improve retrieval via smarter chunking, hybrid search, and reranking. Add a file watcher for fresh indexes, simple evals for trust, and pick lightweight models—setting you up for structured, multi-hop, observable RAG.

Practical, shell-first guide to fixing flaky RAG pipelines: use curl, jq, ripgrep, sqlite3 FTS, strace, and GNU parallel to verify corpus coverage, prove retriever outputs, isolate latency, and validate prompt/context assembly; then guard quality with a tiny recall@k harness and gold queries in CI. Covers common pitfalls (stale indexes, embedding mismatch, bad chunking, filters, truncation) and reproducible bash smoke tests.