Quant / R&D — Trading / Quant

The problem

Polymarket's BTC 5-minute Up/Down market moves too fast for manual trading, and any edge you think you have is worthless until it survives an honest backtest against real market data.

What we built

A four-subsystem quant stack: an LLM-driven strategy-search lab that generates and mutates trading strategies; an ML lab training LightGBM models on 870k historical windows; a live CLOB order-book recorder capturing the market as it happens; and a replay engine that backtests strategies against over 6.3 million historical taker trades.

The result

A research system where every strategy is stress-tested against recorded reality before a single dollar is risked — strategy search, machine learning and honest replay in one loop.

Subsystems in the stack

870k

Training windows (LightGBM)

6.3M

Trades in the replay backtest

Next project

CADEMA

Start a project

Got something worth building?

Book a call

Tell us what you’re working on. We reply to every serious inquiry within a day. · hello@plumbnote.com