Autonomous Trading System

34% returns in 8 months. Fully autonomous. Zero human intervention.

Family Office · Prediction markets · Autonomous · Michael Rosenberg · 2026
34%
Returns in 8 months of live operation
24/7
Fully autonomous operation
Zero
Human intervention required
8 wks
Build time from inception to live

The situation

A family office principal had been trading prediction markets manually for 18 months with strong results – he had developed a systematic approach to identifying mispriced markets and had generated consistent returns. The problem was time: his methodology required constant monitoring, and he was spending 4-6 hours per day on a process that was fundamentally systematic and rule-based. He wanted to automate it entirely.

He had spoken to three technology firms. All three told him it couldn't be done in under six months. We built it in eight weeks.

The system architecture

We designed a three-layer autonomous trading system built on Cloudflare Workers, D1 database, and the Polymarket CLOB API:

Signal Detection Layer: Continuously scans all open prediction markets, calculating implied probability movements, volume anomalies, and cross-market correlations. Applies the principal's systematic criteria – markets where the consensus probability appears structurally mispriced based on public information – to generate ranked trade signals with conviction scores.

Decision Engine: Takes signal output and applies Kelly Criterion position sizing, portfolio concentration limits, and circuit breaker rules. Every trade is sized based on edge strength and current portfolio exposure. The system never risks more than 3% of portfolio on any single position.

Execution Layer: Places orders via the Polymarket CLOB API automatically when all conditions are met. Manages entry, position monitoring, exit execution, and settlement. Handles partial fills, order routing, and slippage management without human involvement.

Risk management

  • Kelly Criterion sizing: position size scales with edge strength, never overconcentrates
  • Portfolio circuit breakers: system pauses if drawdown exceeds 8% in any 30-day period
  • Maximum position limits: no single market can exceed 5% of portfolio
  • Correlation controls: system limits correlated exposure across related markets
  • Daily reporting: automated P&L report, open positions, and risk metrics delivered every morning
  • Full audit trail: every decision, signal score, and trade logged permanently in D1
"I went from 4 hours a day of manual trading to reviewing a one-page daily report. The system is more disciplined than I am – it never second-guesses a signal, never overrides the Kelly sizing, never panic-sells."

The outcome

The system went live in week 8. In the first 8 months of operation, it generated 34% returns with maximum drawdown of 6.2%. It has executed over 800 trades without a single system error. The principal now spends 20 minutes per day reviewing the daily report. The system has operated through multiple high-volatility market events without triggering its circuit breakers.

This engagement represents the frontier of what AI-first operations looks like: a complete, autonomous financial system that operates with the discipline of a systematic hedge fund at a fraction of the traditional cost.

More case studies