S
All posts
Machine Learning 13 minJanuary 25, 2025

Building an AI Trading Signal Engine

Fusing prices, macro data, news and EDGAR fundamentals into an ensemble pipeline — with backtesting discipline at the center.

A trading signal engine is easy to prototype and hard to trust. The difference is discipline — persistence, backtesting and feedback loops, not just a model.

Multi-source ingestion

  • Prices via yfinance.
  • Macro series via FRED.
  • News sentiment scored per ticker.
  • Fundamentals via EDGAR XBRL — 10 years of statements.

From prediction to decision

The /predict pipeline engineers features, runs an ensemble model, and hands the probabilities to a decision engine that emits CALLS / PUTS / HOLD with an assistive advisor.

The part that matters most

Backtesting. Walk-forward, out-of-sample evaluation is what separates a real signal from an overfit curve. Every signal is persisted, outcomes are tracked, and models retrain on a schedule — so the system actually improves.

Lesson

In quantitative ML, rigor beats cleverness. The backtesting harness is the product.

This is a placeholder draft — the full article is coming soon.