Real-time routing & experimentation engine
Every user is scored across 10+ lenders and sent to the one most likely to say yes, instead of walking a fixed funnel.
- 1.8× value per user
- 0 engineering support
Consumer credit, trading systems, government transactions. Four years across three industries where the model, the money and the regulator all have to agree.
I'm a product manager who builds - I write the decision logic, wire up the systems it depends on, and ship the screen the user sees.
Shipped, not specified.
Every user is scored across 10+ lenders and sent to the one most likely to say yes, instead of walking a fixed funnel.
Most Indian applicants have no borrowing history, so the models read bank statements, SMS and app usage too.
When a funnel breaks, these agents read code, logs, events and user records in parallel and come back with the cause.
Every dashboard, pipeline and KPI, built from nothing, then rewritten to cost $25 a day instead of $50K.
A memory layer for AI coding tools: the decisions that live in people's heads, written where any tool can read them.
A multi-model orchestrator on top of Nexus, so each task goes to the model that is actually good at it.
A lending app, a trading floor, and a ministry. In all three, the cost of a wrong decision showed up in someone’s money.
May 2025 - now
Bachatt App · Consumer fintech
Sep 2024 - Apr 2025
Ministry of Finance (DIPAM) · Government
Jul 2022 - Sep 2024
D. E. Shaw · Quant finance
PRDs and prototypes.