FrostGuard AI – Cold Chain

FrostGuard AI – Cold Chain Monitoring & Predictive Rerouting

Overview FrostGuard AI is a real-time cold-chain logistics platform built by Team Red Dragon during a hackathon. It monitors a simulated fleet of medical cargo trucks, predicts temperature breaches before they happen using trained ML models, and recommends the nearest cold-storage facility when a shipment is at risk — all rendered on a live geospatial dashboard.

Core Capabilities

Predictive Breach Detection: An Isolation Forest anomaly detector and a HistGradientBoosting-based forecaster (frost_ml.py, trained via train_model.py) flag abnormal temperature patterns and forecast breach risk 30 seconds ahead. The trained model is serialized once (frostguard_ml.joblib) and loaded at boot rather than retrained on every startup — keeping cold starts fast on resource-limited hosts. Nearest-Facility Rerouting: A K-Nearest Neighbors model (scikit-learn) matches an at-risk truck's GPS position to the nearest of 16 cold-storage nodes across India, so operators get an immediate reroute recommendation rather than raw coordinates. Cached Road Routing: Driving paths between cities are fetched from the OSRM routing engine and precomputed once (precompute_routes.py → routes_cache.json), removing a live external API dependency from the app's boot path. Interactive Dashboard: main_dashboard.py, built with Streamlit, gives operators a live map, fleet metrics, alerts, and a manual failure-injection tool for demoing breach scenarios. Optional Standalone Services: api.py (FastAPI) and Bridge.py (Flask) provide REST endpoints for telemetry ingestion, and Bridge.py includes a Discord webhook integration that posts an alert when a truck goes CRITICAL — opt-in via a DISCORD_WEBHOOK environment variable. These run independently of the main dashboard and aren't required for it to function.

Technologies: Python, Streamlit, scikit-learn (Isolation Forest, HistGradientBoosting, KNN), Docker, REST APIs, OSRM, Discord Webhooks.