Context

A research project born from the curiosity of applying machine learning to forex financial data, with an engineering-first, verifiable approach.

The challenge

The real difficulty isn’t training a model — it’s avoiding the illusion: overfitting, look-ahead bias and optimistic backtests that don’t survive reality.

The solution

I built a feature-engineering pipeline, a correct time-based validation and a rigorous backtesting engine to compare strategies on objective metrics.

The result

A reusable framework to experiment with trading ideas on measurable, honest results — without self-deception.

Stack: Python · Machine Learning · Backtesting