John Sabo
← All work

Systematic trading research platform

Quant Trading Platform Running daily

research platform + dashboard

A research platform for systematic trading. Nine million price rows, 159 strategies through a daily tournament, plus a review team of thirteen agents that audits the work.

What it is

A 22-stage pipeline runs every trading day. It backfills price and filing data, generates signals, runs strategies through a tournament and scores the survivors with GPU models that read news and SEC filings. The dashboard browses portfolios, funds, tickers and signals.

It is also my test bed for agent quality. The Firm, thirteen specialist agents with practitioner doctrine baked in, reviews the platform's output the way a hedge fund desk would: independently, on the record and willing to say no.

The dashboard: 533 daily portfolios, $13.3M under test, market regime and latest signals
The dashboard: 533 daily portfolios, $13.3M under test, market regime and latest signals
The dashboard: 533 daily portfolios, $13.3M under test, market regime and latest signals
The portfolio board: hundreds of strategies, each with its own P&L
The portfolio board: hundreds of strategies, each with its own P&L
The portfolio board: hundreds of strategies, each with its own P&L
The signal log: 500 live signals, filterable by strategy and market
The signal log: 500 live signals, filterable by strategy and market
The signal log: 500 live signals, filterable by strategy and market
A fund view: performance, drawdown and per-fund detail
A fund view: performance, drawdown and per-fund detail
A fund view: performance, drawdown and per-fund detail

Under the hood

What was hard

Keeping the data honest. Backtests are easy to make look good and easy to make lie. Point-in-time flags, accounting audits and out-of-sample discipline are the boring work that makes a number mean anything. The review bench exists because the person who built a strategy is the worst judge of it.

159
strategies
9.25M
price rows
201
commits
2.5 GB
of data and code

Stack

Python PostgreSQL Flask GPU ML (FinBERT) polars

Working on something like this?

One sentence about the problem is enough.

john@johnsabo.org