AI Trading

The Problem

Algorithmic trading tools are either oversimplified Jupyter notebooks or expensive black-box platforms. I wanted an autonomous system that combines technical analysis with LLM-driven decision support — and actually runs in production.

The Solution

Built a fullstack platform where autonomous trading agents execute on configurable intervals, using technical indicators (SMA, EMA, RSI, MACD) and optional LLM confirmation. The system supports three brokers, Bayesian strategy optimization via Optuna, and a dedicated risk gate. Developed in cooperation with Rudiger Hofert and Dario Gibellini at Absolute Software GmbH, Hamburg.

Technical Highlights

• 170+ Python modules across 15+ packages with enforced module boundaries (import-linter)
• 8 LLM providers (OpenAI, Claude, DeepSeek, Qwen, MiniMax, Moonshot, Zhipu) with runtime registry and auto-restore
• Multi-broker abstraction (Interactive Brokers, Alpaca, Mock) with per-account routing via factory pattern
• Autonomous trading agents with state machine lifecycle, APScheduler, and auto-pause after consecutive errors
• Optuna-based backtesting with walk-forward testing, determinism validation, and 5+ built-in strategies
• Risk gate system with asset classification, audit trail, and backtest compliance checks
• AI Goal Orchestration — hierarchical goal decomposition with 30-second tick intervals

Tech Stack

Key Numbers