This is an automated equity trading system I built in Python on the Alpaca API. It runs as a five-stage pipeline, and it runs on paper trading first. Live capital stays held back until the paper record justifies it.
The automation level is a dial, not a switch. There are three operating modes: alert only, semi-automatic with a confirmation step, and full automatic. That way I can turn the autonomy up or down depending on how much I trust the system in a given market.
How it works
- 01Scanner. Filters the universe down to candidates worth a closer look.
- 02Research. Pulls the price and fundamental features each candidate needs.
- 03Prediction. Combines a machine learning model on those features with a sentiment signal.
- 04Risk. Sizes the position and applies the risk controls before anything is placed.
- 05Execution. Places the order through the Alpaca API in the selected operating mode.
Training and infrastructure
The machine learning side is backed by an AI training platform that handles model training and iteration. It uses the Packet API for programmatic training runs, and the whole system is deployed and run on Railway so it can operate continuously rather than only when my laptop is open.
Why the post-mortem matters more than the prediction
Every closed trade runs through an automated post-mortem. The prediction stage gets the attention, but it is the post-mortem that makes the system better over time. A model that is right today and never reviewed is a model that quietly drifts wrong. Writing down what happened, and why, on every trade is the part that compounds.