A self-hosted AI operating system for a venture capital fund
Roots Ventures wanted dealflow, inbox, documents and research in one place, grounded in the firm's own memory and deployed privately, without paying for a dealflow SaaS like Taghash. We are turning their proof of concept into a production system in three phases over eight weeks.
Roots Ventures· Venture CapitalWhat they were running on
- Dealflow spread across email, calendars and shared drives
- Company research repeated for every new deal
- Fund and LP reporting assembled by hand
- A per-seat dealflow SaaS on the shortlist, holding the fund's data on someone else's servers
Why it had to change
A fund's edge is its judgement and its memory. Roots needed AI that could read its inbox, meetings and documents without handing them to a third-party SaaS, and that grounded answers in Indian sources such as MCA and SEBI filings.
The system, on their own server
Inflow OS: email, calendar, Meet and Zoom connected, so every deal is captured automatically
Deal and Research OS: company research grounded in MCA, SEBI, Traxn and Crisil data
Fund, LP and Market OS: portfolio and LP reporting
Self-hosted language models on infrastructure the fund controls
One owned system, so a separate dealflow subscription such as Taghash is not needed
How it went live
Built from the fund's working proof of concept, so the team keeps using what works while each phase goes live.
Related case studies
Sub-10ms execution for a Polymarket trading desk
A real-time risk dashboard for an arbitrage desk, built on its settlement ERP
