Solo developer building AI-powered trading systems — from autonomous factor discovery to local LLM inference.
Three years in, still shipping. Supporter of open-source software and the communities that build it.
Predix is an autonomous AI agent for quantitative EUR/USD forex trading. It automates the full research and development cycle — from factor discovery to backtesting — using a multi-agent LLM framework on 1-minute data.
What makes it different:
- 🧠 Autonomous factor evolution — the agent proposes, codes, and validates its own alpha signals
- 🛡️ Built-in risk management — drawdown protection, cooldown periods, stoploss clustering detection
- 🔄 Walk-forward validation — avoids overfitting across 2020–2026 EUR/USD data
- 🖥️ Real-time dashboard — Streamlit UI for monitoring factor performance and model evolution
- 🔒 134 integration tests — every commit is checked before it lands
Data Flow:
┌──────────────────────────────────┐
│ Qlib Data (1-min EUR/USD) │
│ 2020–2026 · 96 bars/day │
└──────────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────────────┐
│ R&D LOOP (rdagent fin_quant) │
│ │
│ ┌─────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────┐│
│ │ PROPOSE │─▶│ CODING │─▶│ RUNNING │─▶│ FEEDBACK │─▶│ REC││
│ │ (LLM) │ │ (CoSTEER)│ │ (Docker) │ │ (LLM) │ │ ││
│ └─────────┘ └──────────┘ └──────────┘ └──────────┘ └────┘│
│ Bandit sel. LLM evolves Qlib backtest IC/Sharpe/ Pickle │
│ factor|model factor.py in Docker DD metrics session│
└────────────────────────────────────────────────────────────────┘
│ │
┌───────┘ └───────┐
▼ ▼
┌─────────────────┐ ┌─────────────────┐
│ FACTOR TRACK │ │ MODEL TRACK │
│ │ │ │
│ FactorCoSTEER │ │ ModelCoSTEER │
│ FactorRunner │ │ ModelRunner │
│ FactorFeedback │ │ ModelFeedback │
│ │ │ │
│ → result.h5 │ │ → PyTorch preds│
│ IC / Sharpe │ │ + mlflow logs │
└─────────────────┘ └─────────────────┘
│ │
└─────────────┬───────────────────────┘
▼
┌────────────────────────────────────────────────────────────────┐
│ STRATEGY GENERATION PIPELINE │
│ │
│ Load top factors ──▶ LLM strategy code ──▶ OHLCV backtest │
│ │
│ Optuna: Stage 1 (10) → Stage 2 (15) → Stage 3 (5) trials │
│ Accept: Sharpe ≥ 1.5 · DD ≥ −0.30 · WR ≥ 0.40 │
└────────────────────────────────────────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────────────┐
│ PORTFOLIO OPTIMIZATION │
│ Mean-Variance / Risk Parity / Black-Litterman │
└────────────────────────────────────────────────────────────────┘
│
▼
┌────────────────────────────────────────────────────────────────┐
│ LIVE TRADING (closed-source) │
│ ftmo_live_trader.py · FTMO signals │
└────────────────────────────────────────────────────────────────┘
Autonomous Trading Agents — Multi-agent LLM frameworks that discover, evolve, and validate trading strategies end-to-end
Local LLM Integration — Running AI systems fully offline with llama.cpp (no cloud dependency)
Open-Source Tools — Pine Script strategies and Python frameworks for the trading community
Full Trading Pipelines — From raw kline data to live execution, built and maintained independently
Core & AI
Data & Finance
Local LLM & Inference
UI & Infra
| Project | Contribution |
|---|---|
| TradingAgents ⭐ 34k | Added llama.cpp local LLM support — run multi-agent stock analysis fully offline via .env config |
| OpenStock ⭐ 9.9k | Updated deps, fixed Inngest v4 API, force-dynamic for auth routes — resolved 28 vulnerabilities, migrated Inngest v3→v4 |
Premium models & collaborations → tpt.requests@pm.me
Mastodon → @TPTBusiness@mastodon.social
⚠️ All content is for educational purposes only. Past performance does not guarantee future results.