Quantitative research lab · Nordhorn, Germany

Systematic trading, engineered with AI.

Aureus Capital builds and stress-tests automated trading systems for MetaTrader 5. AI agents help us write, review and audit strategy code. Every decision to trade passes through hard, deterministic risk rules.

See our projects Get in touch

XAUUSD · H4Donchian 20breakout signal

Simulated data, for illustration only

MQL5 · PythonStrategy and tooling code

Walk-forward · OOSRobustness before any live run

Multi-agentClaude-assisted code review

Self-fundedIndependent since 2026

Research pipeline

From hypothesis to forward test

Each strategy moves through the same six stages. A system that fails a stage goes back, or gets dropped.

Hypothesis

A market behaviour we can describe precisely: a breakout, a session pattern, a volatility regime.

notes · TradingView

Implementation

The rule set becomes an Expert Advisor or a Python model, with inputs exposed for testing.

MQL5 · Python

Backtest & optimisation

Tick-based tests in the MT5 Strategy Tester; optimisation runs across thousands of parameter sets.

MT5 tester · .opt files

Robustness

Walk-forward and out-of-sample checks, parameter stability maps and Monte Carlo resampling to catch overfitting.

WFA · OOS · Monte Carlo

Risk layer

Position sizing, daily loss limits and drawdown stops live in fixed code, outside any model.

deterministic rules

Forward test

The system runs on demo accounts under live conditions before it is considered finished.

demo · monitoring

How we use Claude

AI does the research work. Rules make the trades.

Language models are excellent at reading code, reports and logs. We use them for that, and keep them away from the order button.

Agent output informs research only. Orders pass through fixed code, never through a model.

Code
Expert Advisor reviewSeveral agents read the same MQL5 source independently and flag look-ahead bias, unsafe order handling and risk-rule violations.
Reports
Backtest forensicsClaude reads Strategy Tester reports, deal exports and optimisation results and checks them against our robustness criteria.
Build
Tooling with Claude CodeOur analysis software, parsers and data pipelines are developed with Claude Code as a pair programmer.
Research
Event and news researchMatching news flow to prediction-market questions and summarising what changed, for a human to judge.
Limits
What AI does not doNo model places orders, sizes positions or overrides a stop. Those rules are written, tested and versioned by hand.

Projects

What we are building

PyQt6 · desktop

EA Forensics Pro

A desktop application that analyses Expert Advisors: a risk detector, a grade calculator and a parser for binary MT5 optimisation files, tested on 13,994 optimisation passes.

PythonPyQt6.opt parser

MQL5 · MT5

Expert Advisor portfolio

Rule-based systems for gold and indices: fractal breakouts, Ichimoku-based trend logic and session strategies, each run through the full pipeline.

XAUUSDPaliKijunBobyStarTreck2GoldForecast

TradingView · BTC H4

Aureus Direction

A custom Donchian-channel indicator for Bitcoin on the four-hour chart, used as a directional filter for discretionary trading.

Pine ScriptDonchianBTC

Python · alerts

PolyEdge

A research bot that matches incoming news with open Polymarket questions and sends alerts when an event may move a market.

NLPPolymarketClaude API

Evidence over backtests

A good equity curve is a starting point. We trust results that survive out-of-sample data and parameter changes.

Risk is code

Limits are hard-coded and tested like any other feature. Nothing about risk is left to judgement in the moment.

Research, not advice

We build software and research for our own use. We do not manage client money or give investment advice.

Contact

Let's talk research.

roman@aureuscapital-ai.de

Questions about our tools, collaboration on research, or partnerships with AI and data providers.