What it takes to build the platform behind a quant fund
Before Meta I co-founded Theme Analytics, a quantitative hedge fund, and built the platform it ran on. It was acquired in 2025. People hear "quant fund" and picture the strategy -- the signal, the edge, the thing that makes money. That part was real, and I worked on it. But an equal amount of the work was the infrastructure underneath, and the fund needed both in equal measure.
Half the job is infrastructure. Before any idea earns a dollar, you need clean prices and fundamentals, corporate-action adjustments, borrow and short-locate availability, a backtester you can actually trust, a portfolio optimizer that respects real constraints, execution, reconciliation, and daily P&L that ties out to the penny. None of that is the edge. All of it has to be right before the edge means anything.
The platform. Ingestion pipelines pulled from several market-data sources and normalized them into one consistent store. A feature layer turned that raw data into the inputs research actually used. A backtester ran strategies over history net of costs. A constrained optimizer turned target signals into a tradable portfolio under position and risk limits. Then execution, reconciliation, and reporting closed the loop each day. It was Python and Postgres on AWS, the daily flow orchestrated with Dagster, the infrastructure itself in Terraform so the whole stack could be stood back up from code.
The hard part was trust, not features. Adding a data source or a new report is easy. Making the numbers honest is not. A backtest is only worth running if you believe it, and most of the engineering went into the things that quietly make backtests lie: survivorship bias, look-ahead bias, costs and slippage modeled too optimistically, a corporate action applied a day late. A backtester that flatters you is worse than none, because it gives you confidence to lose money. Most of my job was making ours refuse to flatter anyone.
The platform was a multiplier, not a substitute. I worked both sides, the platform and the research. The platform was the HOW; the research -- the reason a price was wrong -- was the WHAT, and the two took roughly equal effort. What good infrastructure does is let a real idea be tested honestly and run at size without falling over. It raises the ceiling on a good thesis and does nothing for a bad one. That distinction has shaped how I think about building ever since, including how I use LLMs now: they are extraordinary at the HOW and no help at all with the WHAT.
The platform turned into a sales tool. It was good enough that we started showing it live in pitches. I'd pull it up on my phone at a networking event and walk an investor through real positions and live P&L, or screenshare it during a Zoom pitch, and it built more confidence than any deck could. One prospective investor liked it enough to ask whether they could license the software itself, so we walked out of a single meeting with both their allocation and a paying SaaS customer.
Theme was acquired in 2025 and the platform now runs under new owners, licensed to other managers. The fund was the reason to build it, but the part I'm proudest of is the boring part: an honest pipeline that other people could trust their money to.
← Writing