Devesh Joshi

Vantedge AI (Y Combinator W22), AI infrastructure for investment funds

• Own the client-facing half of a deployment: discovery, project logic, coordination across the delivery team, and the daily relationship. Deployed into two live funds at the same time.
• Main point of contact for the Managing Partner of a $200M private credit fund, on a standing daily call. He builds his own tools and benchmarked us against the best AI products throughout. He signed before his own fund had closed, used the product every day, and became our largest source of referrals.
• Designed a maker-checker system for work where accuracy matters: one model does the job, a second model checks it, Python handles all arithmetic, and every figure links back to the exact spot on the scanned page it came from. 100% on benchmark testing.
• Set the logic for a multi-stage document extraction pipeline, so financial statements in wildly different formats come out standardised. Paired it with an engine that fills any Excel template from a data room of hundreds of files.
• Won back a CRM purchase already in motion by specifying and shipping an LP CRM to the client's own spec, including one-click export of every deal and attachment to answer their lock-in worry.
• Worked inside the fund's confidentiality rules. Investor performance data stayed on their systems, only NDA-covered deal data entered our cloud.

Eight Capital, early-stage venture fund (~$30M AUM), affiliate of Vantedge AI

• Evaluate early-stage deals with a focus on backing Y Combinator companies before Demo Day, and own the diligence and reporting systems behind them.
• Run founder diligence calls across the pipeline and feed comparative scoring into investment and follow-on decisions.
• Built a founder scoring framework rating pedigree, past startup experience, domain expertise, team make-up and risk, plus clean datasets for priced rounds and SAFEs.
• Automated investment and IC memo writing. Preparation dropped from about six analyst-days a week to one, and the memos came out more consistent.
• Rebuilt recurring fund reporting including revenue updates and NAV calculations. Reconciliation went from a full day to about an hour, with fewer manual errors.

Digixpressions Media, led a team of six building a retail investor research product

Morningstar, dilutive securities across US, Canadian and Australian reporting standards

• Held 98.5% data quality for three consecutive months. Taught myself to code on this job, because the manual version of it was unbearable.