One queryable company brain across 12 years of deals.

Scope

Build an internal AI knowledge system that ingests the firm's full historical archive — IC memos, market research, broker emails, LP updates, partnership documents — and lets the investment, asset management, and IR teams ask questions and get cited answers in seconds.

• 12 years of deal history, market views, and partner judgment sat across email, shared drives, and a few partners' memories. New hires were taking 6+ months to be useful on deal discussions.
• Every LP question — 'have you ever done a deal like this before?', 'what was your view on this submarket in 2021?' — required pinging multiple senior people who were already over-allocated.
• Important context from past deals was being lost as the team grew. Decisions were being made without the firm's own history in the room.

• Mapped the firm's existing knowledge sources with the partners and IR lead, then prioritized by what gets asked most in IC and LP conversations.
• Designed a workflow that ingests, deduplicates, and indexes documents — with strict access controls per matter type and a clear separation between firm-internal and LP-facing material.
• Built a chat surface that answers in plain English with citations back to the source document, so users can verify any claim in one click.
• Deployed in stages: first to partners only, then to the IC and IR team, with structured feedback loops to improve citation quality and add new document sources.

Measured after the first 90 days.