Inspiration
Current wearables operate in the past. They analyze yesterday's data to tell you that you are exhausted today—when the biological damage is already done. For tactical operators and high-performance athletes, reactive data is useless. We needed a forward-looking radar for the autonomic nervous system to intercept overtraining before it happens.
What it does
Aegis predicts a user's Heart Rate Variability (RMSSD) 24 hours into the future. It ingests 7-day rolling biometric telemetry, forecasts the autonomic state, and calculates the variance against an established physiological baseline. Based on strict mathematical thresholds, it issues deterministic Red, Yellow, or Green training protocols.
How we built it
The core inference engine is a Gradient Boosting Regressor trained on deterministically engineered physiological data. To prevent black-box medical outputs, we integrated SHAP (SHapley Additive exPlanations) game theory, which dynamically isolates the exact biometric vector (e.g., compounding sleep debt vs. acute physical load) dragging the nervous system down. The frontend is deployed via Streamlit for rapid CSV ingestion and visualization.
Challenges we ran into
Handling the physics of shifting physiological baselines. Initially, our clinical logic used a rolling 7-day average as the baseline. This created a "shifting goalpost" flaw where a severe 7-day systemic crash lowered the average so much that further degradation registered as a false positive "recovery." We had to fundamentally re-architect the mathematical inference layer to anchor predictions against a strict, unmoving demographic baseline constant to ensure accurate clinical variance triggering.
