All case studies
Lumen LabsAI2025

RAG-powered research, at enterprise scale.

We built a private knowledge engine that ingests 12 file formats, cites sources, and answers across 200K internal documents.

LangChainpgvectorOpenAIPython
L
Lumen Researchmodel · v4
Summarize Q3 risks across all internal memos.
Across 47 memos, Q3 risk concentrates in: supply-chain (38%), regulatory (22%), and FX (18%). 
memo-0142.pdfq3-strategy.mdops-review.docx
Ask anything across 200K docs…
Challenge

Lumen's analysts spent hours hunting through Confluence, SharePoint, and Slack to answer the same research questions. Hallucination-prone consumer tools were a non-starter.

Approach
  1. 01Designed an ingestion pipeline for 12 file formats with provenance.
  2. 02Built a hybrid retrieval stack (BM25 + pgvector) tuned per domain.
  3. 03Wrapped responses in a citation contract — every claim links to a source.
  4. 04Rolled out behind SSO with row-level access controls.
Outcome
3M / mo
Queries answered
200K
Documents indexed
94%
Citation accuracy
Working on something similar?

Let's talk about your project.

A 30-minute call. No pitch. We'll listen, ask sharp questions, and tell you whether we're the right fit.

Start a conversation