Enterprise RAG: fixing retrieval first cuts failed retrievals by 67% before touching the model
LangChain · x · 2026-11-01
Kacper Klusek of Soba Labs published a LangChain community guide arguing that enterprise RAG failures usually sit in retrieval, not the model: answers get badly chunked, embedded without context, outranked by near-duplicates, or never indexed.
The recommended stack, ordered by ROI, builds on Anthropic's published evaluation (baseline 5.7% top-20 failure rate):
- Contextual embeddings: 3.7%, 35% fewer failures
- + contextual BM25 hybrid search: 2.9%, 49% fewer
- + reranker (retrieve top 150, pass top 20): 1.9%, 67% fewer failures — more than any model swap buys
He also stresses LangSmith online evals to catch the silent retrieval decay that starts when documents change, plus a pre-ship checklist for agentic RAG builds.
More from coding & agent
- Blogger lets Grok spend up to $100 autonomously and receives a LEGO Harry Potter gift — venturetwins · 2026-11-01
- TensorAgent Adds Local Multimodal Indexing for Documents, Images, Audio and Video — fuzhongkai · 2026-11-01
- Theo claims new model hallucinates 10x worse in real codebases; bugs traced to ACP integration — cgarciae88 · 2026-11-01
- One prompt to an AI agent adds note-taking and search to X bookmarks — HaijunXia · 2026-11-01
- Running 16 desktop AI agents at once on a single Mac, clicking via accessibility refs — cramforce · 2026-11-01
- Grok helps user find 74GB Cursor chat backup, doubles free disk space — prasenx · 2026-11-01