Built a quantitative research desktop pipeline that converts OHLCV data into causal SMC market-structure events, custom context/Volume Profile features and TP/SL-labelled datasets, then trains XGBoost with chronological train/validation/test splits, calibration, validation-only thresholding and leakage checks. The research layer is evaluated with realistic backtesting including fees, slippage, overlap control and equity risk sizing. The public demo preserves the architecture but replaces proprietary scoring, targets and features with generic examples.