← Back to Projects
MLResearch2025Shipped

OSAS Detection System

End-to-end ML pipeline for multimodal physiological time-series classification with production-ready training infrastructure.

Technology Stack

PythonPyTorchTransformersNumPyPandasTime-Series AnalysisHealthcare AI

What I Built

  • Designed multimodal transformer architecture combining vital signs, ECG/PPG, and PSG signals
  • Built complete training pipeline with checkpointing, logging, and macro-F1 optimization
  • Implemented patient-aware data splitting to prevent data leakage in clinical datasets
  • Developed comprehensive preprocessing and EDA tooling for exploratory analysis
  • Applied attention mechanisms and positional encoding for long-range temporal dependencies