← All work
Agentic System
Autonomous Research Engine
A recursive harness that continuously generates hypotheses, backtests them against live market data, and surfaces ranked signals — unattended.
Runs 24/7 on dedicated hardwareVisit yacht.bot ↗
The problem
Finding edge in prediction markets means testing an endless space of strategies against constantly-moving data — far more than a person can evaluate by hand.
The approach
I ported a recursive autoresearch pattern into a harness that proposes strategies, backtests them against real market history, keeps what survives, and feeds the findings back into the next round — running continuously on a dedicated machine.
What I built
- A recursive research loop that proposes and self-evaluates strategies
- A backtesting engine validating against live market history
- A monitoring dashboard surfacing ranked, hold-to-settlement signals
The outcome
An always-on research engine that does the work of a quant research desk on a single dedicated node — and surfaced that a maker, hold-to-settlement strategy beat the naive live approach.
Stack
PythonRecursive agent loopBacktesting engineNext.js dashboard