John Jayasankar

John Jayasankar

I build AI agents for high-stakes financial workflows.


2025 to present
Quantile Technologies

I am Lead Product Manager for Portfolio Optimization at Quantile Technologies (an LSEG business). I ship production AI agents and a reusable human-supervised architecture across enterprise financial workflows. A compression setup agent collapsed a 3.5-hour expert setup bottleneck to 8 minutes and scaled run volume 3× at constant headcount. An incident triage agent returned 750+ senior engineering hours a year by diagnosing incidents before they escalated. The agent platform made context, tools, actions, and evaluation reusable across five enterprise agent systems.

2023 to 2025
Quantile Technologies

I was Product Manager for Portfolio Optimization at Quantile. I owned discovery, strategy, launch, and GTM for optimization products used by global banks, and the work generated $XM+ in new and expansion ARR across rates, FX, and cross-currency. That meant 0→1 launches in multilateral compression, valuation/validation, and FX.

A few of the systems from this stretch: Multilateral cross-currency compression made $X.XT of previously ineligible bilateral notional available across 12 currency pairs. Dual-source valuation surfaced valuation discrepancies 48 hours earlier and cut failed-run resubmissions 91%. FX forward and NDF compression embedded margin intelligence in the optimizer and hit 100% proposal acceptance across 40+ live runs.

2022 to 2023
OpenGamma

I was a Product Analyst at OpenGamma, later acquired by Trading Technologies. I originated and shipped a 0→1 pre-trade margin simulator used in front-office workflows. I led discovery across trading, treasury, risk, and operations, with adoption across 20+ enterprise clients. The point was simple: capital should be a decision before you execute, not a surprise after.

2021
Wells Fargo

Summer Analyst on the FX / Rates trading desk at Wells Fargo. This is where I learned that in markets, a late failure is not an ops inconvenience. It is the product.

2021
Wharton School

Research Assistant at the Wharton School, working on reporting quality and incentives. Useful training for caring about whether a number is actually true before anyone acts on it.

Earlier
Hartford Funds

Summer Intern at Hartford Funds on Mutual Funds & ETFs. Early market-structure reps before I moved fully into product.

Earlier
Ocean Trail Partners

Private Equity Summer Intern at Ocean Trail Partners. Early diligence reps.

2018 to 2022
Haverford College

B.A. in Economics & Linguistics at Haverford College. GPA 3.9, cum laude, High Honors in Linguistics. I liked the combination: models of behavior on one side, precision about language and meaning on the other. That pairing still shows up in how I design agent boundaries.

bio

John Jayasankar is a Lead Product Manager in New York. He builds production AI agents and 0→1 financial infrastructure for complex, high-stakes workflows. At Quantile (LSEG) he has shipped agents and optimization products used by global banks. Previously he originated a pre-trade margin simulator at OpenGamma. Independently he built and shipped RideLens, Daylight, RailDrop, and Gridiron, collected on Labs.

selected systems

how I build

Agents should earn autonomy. I start with the workflow, then give the system only the context, tools, and typed actions it needs, with a human gate on anything consequential. Model capability is one layer. Autonomy grows one rung at a time, on evidence. More on the control model.

featured writing

Short theses for now. Longer notes land on Substack.

independent products

Four of the eight independent products have case writeups here. All of them are shipped live and collected on Labs.

RideLens is every ride, one comparison. Live roads. Real rate cards. Uber, Lyft, Empower, and Curb ranked before you book. I built it end-to-end because I was tired of opening four apps and guessing. RideLens case writeup.
Daylight is adaptive display lighting for macOS: warmth and brightness on a schedule you set, entirely offline, with a printed six-layer precedence ladder and three levels of certainty about whether a change actually took. Daylight case writeup.
RailDrop watches Amtrak listed fares across your travel window and emails only when a qualifying option improves. It fails honestly when the source is down and never invents fees, fares, or fake itineraries. RailDrop case writeup.
Gridiron follows every live NFL and college football game on its own 3D field, with the reported ball spot, win probability, and odds from named sources. It draws only what a provider reported, and says so when something is missing. Gridiron case writeup.

selected outcomes

Operations agent that collapsed a 3.5-hour bottleneck
Quantile · 2025
Setup 3.5h → 8m · 3× run volume · 0 AI config errors / 6 mo
Incident agent that returned 750+ engineering hours a year
Quantile · 2025
Investigation 4.5h → 11m · −78% escalations · domain MCP
Multilateral compression for trillions in trapped notional
Quantile · 2024
18 banks · 12 currency pairs · +34% notional reduction / run
Dual-source valuation that caught discrepancies 48 hours earlier
Quantile · 2024
48h earlier detection · −91% resubmissions · 24 banks
FX compression with margin intelligence in the optimizer
Quantile · 2024
40+ live runs · 100% proposal acceptance · −94% live failures
Pre-trade simulator that made capital a decision, not a surprise
OpenGamma · 2022 to 2023
0→1 front-office product · adoption across 20+ enterprise clients

misc