RAG Knowledge Base over PDFs and Confluence
Your internal docs become a searchable, cite-able knowledge base. Claude answers questions with exact passages highlighted.
How it works
1. Ingest from anywhere
Drag-drop PDFs plus OAuth into Notion and Confluence. Unstructured.io handles parsing; content is chunked and embedded.
2. Hybrid retrieval
Semantic (pgvector) + BM25 keyword search with Cohere rerank on top — so obscure acronyms and numbered clauses still win.
3. Answer with citations
Chat UI returns answers with inline citation links that deep-link to the exact source paragraph. Admin panel exposes failed queries.
60-second Loom
45-second flow: ingest 30 PDFs → ask a policy question → see the answer with 3 citations → click a citation → jump to the exact paragraph.
Get the full build spec PDF
One-page PDF with the architecture, the tech stack, the timeline (2–3 weeks), and the investment range ($9,990 – $25,000). Sent to your inbox in under a minute.
Who it’s for
Legal, scientific R&D, and operations teams with 500+ internal documents who need a cite-anchored answer surface that employees actually trust.
Tech under the hood
Frequently asked questions
How is this different from ChatGPT with file upload?
Hybrid retrieval (semantic + BM25 keyword) plus a Cohere rerank step means domain-specific acronyms, numbered clauses, and low-frequency terms still surface. Citations deep-link to the exact paragraph — not a whole PDF.
What sources can you ingest?
PDF (scanned + text), Markdown, HTML, DOCX, CSV, Confluence, Notion, Google Drive, Dropbox, S3, and any HTTP(S) endpoint. Incremental re-indexing keeps changes fresh.
Is it secure for internal docs?
Yes. Deployable in your VPC or ours. Row-level security via Supabase RLS ensures users only see content they have source-system permissions for. SOC 2 path available.
How much does it cost to run?
For 10k documents and 5k queries/month: ~$90 for pgvector hosting, ~$45 for Claude inference with prompt caching, ~$25 for reranking — under $200 all-in on typical usage.
Want one of these in your business?
We’ll scope, build, and hand off a production-ready system in 3–4 weeks.