Build a RAG prototype that answers from your docs, with sources

Most teams have a wiki, or a Notion, or a folder of SOPs that almost no one reads, probably because finding anything in there is painful. RAG can fix this. Done badly, it just returns plausible-sounding nonsense and your team stops trusting it in one shot (pun intended).

What I'll do in 5 days is build you a working RAG prototype over up to 200 pages of your content. Your team gets an AI assistant they can poke at, the system shows citations with every answer, and you have evidence one way or the other on whether it's worth scaling to your full document set.

The timeline of it:
- Day 1: ingest your content (Notion, Drive, PDFs, Markdown...most formats work)
- Day 2-3: build the retrieval pipeline with embeddings and reranking, not naive vector search that drowns in noise
- Day 4: chat interface deployed somewhere usable (basic web widget or Slack bot)
- Day 5: evaluate on 10 test queries you give me, walkthrough call

This is the "should we roll this out to the whole team / customer base?" version. If it proves out, scaling to a bigger corpus, more sources, a custom UI, or production monitoring is a separate conversation.

For context: I run a custom RAG over my own knowledge base. A few thousand notes that grow daily, indexed across multiple formats. The ways these systems quietly fail in production (stale content, retrieval noise, citation drift) are already on my checklist because I've debugged most of them on my own setup.