From raw data to executed order
A transparent look at the six-step pipeline that drives every signal.
Visual pipeline
Signal generation to execution
01
Market data
Bars, spreads, microstructure features
02
Model layer
XGBoost inference + thresholding
03
Signal relay
Authenticated WebSocket delivery
04
Local execution
Orders routed from your IB session
Data ingestion
The pipeline pulls OHLCV bars from Interactive Brokers historical data API for MNQ, MES, NQ, and ES futures. Bars are stored in the cloud and updated in real time during market hours.
Feature engineering
Hundreds of technical indicators — momentum, volatility, volume profile, inter-contract spreads, and time-of-day seasonality — are derived from raw OHLCV data. Feature computation is cached and incrementally updated each bar to minimise latency.
XGBoost inference
A gradient-boosted classifier trained with Optuna hyperparameter optimisation outputs per-bar probability scores (P↑, P↓). Walk-forward validation prevents look-ahead bias. The model re-trains periodically on an expanding window.
Signal & position sizing
Raw probabilities are converted to directional positions (Long / Flat / Short) using a Sharpe-optimal threshold tuned on the validation fold. Position size is scaled by a confidence-accuracy matrix derived from held-out validation data.
WebSocket delivery
Within milliseconds of each bar close, the signal payload (contract, direction, probability, threshold) is published to our cloud WebSocket server and pushed to all active subscriber sessions.
Local order execution
The AureonQuant desktop client receives the signal and forwards it to your locally-running IB Gateway via the IB API (port 7497). Limit orders are placed in your own account. Fill confirmations are sent back to the server for performance tracking.