Projects

Quantitative engineering · Private

Bandit Thompson

My algorithmic trading system — years of research treated as a lab, not a hunch.

Bandit Thompson trading XAUUSD M15 on a live account (cTrader).

What it is

A bot that trades gold, on the scientific method

A bot that trades gold (XAUUSD) on the 15-minute timeframe, in production on a live account. The premise: the market is mostly noise — so every decision the bot makes is a hypothesis, tested against that noise, and only stays in the system if the data proves a real edge. Nothing gets in on intuition.

The discipline

Infrastructure to separate luck from edge

Deterministic backtests (same input → same output, 100%), out-of-sample validation, and a notebook of documented findings that govern what stays and what goes. Built in C#/cAlgo on cTrader, with an external Python system for optimization (Optuna).

The learning

I killed the piece the bot is named after

The original version used a Thompson Sampling algorithm — the “bandit.” When the analysis showed it added no edge at all, I removed it — even though it's the project's name. What remained (market-structure detection + regime via HMM) came out about 48% smaller in code, faster and stronger. I kept the name as a lesson: in the scientific method, you kill your darlings when the data says so.

If you got this far, send a message.

Even if just out of curiosity. I reply on LinkedIn or by email.