QQuestion

What's your RAG-for-PM stack for internal docs Q&A?

MP@minji· Senior PM at Toss· 🇰🇷 KR· 1d ago

We have ~3k Notion docs and want PMs to ask "Has anyone shipped X before?" and get cited answers. Trying to decide between LangChain + Pinecone vs Vercel AI SDK + a hosted vector DB. What's your team using?

#rag#vector-db#internal-tools

◐ 3 Answers

✓ BEST ANSWER

@sarah · Product Manager at Stripe · 🇺🇸 US · B2B SaaS2,341 · 4h ago

We went with Vercel AI SDK + Turbopuffer. The simpler the better — Pinecone is overkill at 3k docs. Set up the retriever as a tool and let the model decide when to fetch. Took half a day end-to-end and the latency is great.

@kenji · Group PM at Fintech Co · 🇯🇵 JP · Fintech890 · 6h ago

We landed on LangChain + Pinecone but mostly because our infra team standardized on Pinecone. If starting fresh today, I'd start simpler and only graduate when you actually feel the pain.

@yuki · Product Manager at Mercari · 🇯🇵 JP · Marketplace410 · 1d ago

Tip: hashing on document section, not page, made retrieval way more relevant in our case. Pages were too coarse and we ended up with cited chunks that didn't really answer the question.