[Show and Tell] Lumina Capital — Autonomous Multi-Agent AI Hedge Fund using TimesFM for Zero-Shot Probabilistic Stock Forecasting

[ramuza0113-sketch]

Hi TimesFM community and Google Research team! 👋

I wanted to share an open-source project where we integrated Google TimesFM into a production-grade algorithmic trading pipeline:

Lumina Capital (AI Hedge Fund Engine v6.0)
🐙 GitHub: https://github.com/ramuza0113-sketch/lumina-ai-hedge-fund
🚀 Product Hunt: https://www.producthunt.com/posts/lumina-capital

---

📈 How We Leverage TimesFM

In financial markets, traditional statistical models struggle with regime shifts. We integrated TimesFM as our Zero-Shot Probabilistic Forecaster:

1. Intraday & Swing Horizon Forecasting: Generating quantile price paths (10th, 50th, 90th percentiles) across multi-asset universes (US & TSE equities).
2. Ensemble with Chronos & Qlib Alpha158: TimesFM's zero-shot predictions are fed alongside Microsoft Qlib's 158 mathematical factors into a LightGBM gatekeeper to calculate probability-weighted entry confidence.
3. LangGraph Multi-Agent Feed: The forecast distribution is injected directly into our 8-legend investor council (Buffett, Burry, Soros personas) to evaluate risk-reward skewness before order execution.

Huge thanks to the Google Research team for open-sourcing TimesFM! We'd love to hear your feedback, thoughts on time-series scaling, or ideas for financial domain adaptation.