BREATHE SAFE

Air pollution in Delhi NCR regularly crosses hazardous levels, but most residents only find out after exposure — existing apps show current AQI with no reliable forecast, no zone-level detail, and no proactive health guidance. BreathSafe solves this by combining a trained forecasting model with a practical, everyday-usable interface. At its core, BreathSafe uses a Temporal Fusion Transformer (TFT) model trained on historical pollution and weather data for Delhi NCR, served through a FastAPI backend with endpoints for prediction, health checks, city-level AQI indexing, and nearby safe-zone lookup. We benchmarked the TFT model against Persistence and 7-Day Simple Moving Average baselines during development to validate real forecast value rather than shipping unverified predictions. On the frontend, users get an interactive map (built with Leaflet.js) showing live AQI by zone with color-coded severity, historical trend and pollutant-breakdown charts (Chart.js) for PM2.5, PM10, and NO2, and a forecast panel with multi-step predictions. Beyond raw numbers, BreathSafe translates AQI into everyday decisions: health advisories per AQI band, mask recommendations, and exercise-safety guidance — so a parent or an asthma patient gets an actionable answer, not just a number. Our target users are residents of high-pollution NCR zones, especially vulnerable groups (children, elderly, asthma/respiratory patients), schools deciding on outdoor activity, and local administrators needing zone-level pollution visibility. The app is built to work as both a native Android app and a lightweight web app, so it's accessible without requiring an install — important for quick adoption during high-pollution emergency periods. What makes BreathSafe different from a typical AQI-display app is that it's forecast-first, not just monitoring-first: the goal is to tell people what air quality will be in the next 24 hours to a week, not just what it is right now, giving them time to act — plan indoor activities, get a mask ready, or avoid a specific route or zone. We also built the system to be transparent about model confidence rather than presenting predictions as guaranteed fact. The project originated from our team INNOVERSE's work toward SIH 2026 problem statement SIH26082 (Air Pollution–Weather Coupled Forecasting System, Delhi NCR focus, under MoES), and we've continued building it out as a fuller platform: real-time map, historical trend analysis, alerting, and a tested, phased backend/frontend rollout with regression testing at every stage to ensure stability as features were added. BreathSafe is efficient by design — it reuses one FastAPI backend for both the native app and web frontend, keeps the map/chart/forecast UI lightweight (no heavy new frameworks), and was built with rapid, test-verified iterations so each new feature phase was confirmed working before the next was layered on.