← All projects

ML Forecasting & Automated Trading Decision Engine

2023 – 2026

A hybrid ML + rules-based system for forecasting energy prices and safely automating battery trading decisions.

PythonTransformer modelsBacktestingDrift monitoring

Impact

  • Transformer-based forecasting for energy prices and battery health, retrained weekly
  • Rule-based risk controls (circuit breakers, position limits) layered on top of ML predictions
  • Benchmarked 15–25% energy efficiency gains and ~30% extended battery lifespan across pilot deployments

Automating financial decisions on top of an ML model is only safe if you don’t trust the model blindly. This project paired a forecasting model with hard-coded guardrails so the system could act autonomously without taking on unbounded risk.

The problem

The platform needed to make real-time trading decisions for battery storage operators — when to charge, when to discharge, when to sell back to the grid — based on forecasted energy prices and battery health. Getting this wrong isn’t just a bad prediction, it’s a financial and hardware risk.

What I built

The core is a transformer-based forecasting model predicting energy prices and battery health, retrained weekly against fresh market data. Rather than letting the model act unsupervised, I combined it with a rule-based risk layer — circuit breakers and position limits — that constrains what the automated system is allowed to do, regardless of what the model predicts.

Before any production rollout, I backtested the forecasting models against historical market data, and kept continuous retraining and drift monitoring running in production to catch model degradation early.