AeroHealth AI

AeroHealth AI turns live ward AQI into instant, multilingual voice warnings via Gemini AI to protect vulnerable citizens in seconds.

Comment

Inspiration

Urban air spikes trigger acute respiratory emergencies every year. While high-precision monitoring stations capture toxic particulate spikes in real time, official public warnings arrive hours late, remain generic, and are published as text-heavy bulletins in English.

To an asthmatic child or a non-literate laborer, dangerous particulate thresholds:

$$\text{PM}_{2.5} \ge 250\ \mu\text{g/m}^3$$

are completely meaningless. We were inspired to turn raw atmospheric telemetry into instant, actionable, mother-tongue guidance for vulnerable citizens.

What it does

AeroHealth AI connects live air monitoring directly with automated civic protection:

  • Ward-Level Mapping: Visualizes real-time AQI overlays on an interactive municipal ward map, pinpointing schools, hospitals, and eldercare facilities.
  • Contextual Health Guidance: Gemini AI translates raw pollutant data into clinical, demographic-tailored precautions for students, seniors, and outdoor laborers.
  • 1-Click Multilingual Voice Dispatch: City authorities can review AI recommendations on a dashboard and push localized voice advisories and notifications directly to citizens' phones in under 30 seconds.

How we built it

  • Frontend & GIS: Built with React and Leaflet.js using OpenStreetMap layers to render ward polygons and critical infrastructure markers.
  • Sensor Stream: Ingests live ambient pollutant metrics (PM₂.₅, PM₁₀, NO₂, O₃) via the WAQI API.
  • Risk Engine: Computes a ward-level vulnerability index accounting for infrastructure density:

$$R_w = \left(\frac{\text{AQI}w}{\text{AQI}{\text{threshold}}}\right) \times \left(1 + \sum_{k} \omega_k \frac{N_{k, w}}{A_w}\right)$$

  • Intelligence & Voice: The Gemini API generates structured, localized advisory scripts, which Google Cloud Text-to-Speech converts into natural regional audio broadcasts.
  • Distribution: Managed through Firebase Cloud Messaging (FCM) topic-based pub/sub channels for instant delivery.

Challenges we ran into

  • Spatial Interpolation: Sensor stations rarely align with administrative ward polygons; we implemented inverse-distance-weighted spatial interpolation to calculate representative ward averages in real time.
  • Zero-Hallucination Guardrails: Health advisories cannot afford clinical ambiguity; we engineered strict JSON schemas and system prompts in Gemini to ensure deterministic medical recommendations.
  • Low-Latency Audio Streaming: Delivering synchronous multilingual audio on the fly risked pipeline delays; we introduced asynchronous voice synthesis batching to keep the full dispatch cycle under 30 seconds:

$$T_{\text{dispatch}} = T_{\text{infer}} + T_{\text{TTS}} + T_{\text{FCM}} < 30\text{ s}$$

Accomplishments that we're proud of

  • 30-Second Emergency Pipeline: Reduced the civic delay between toxic spike detection and public broadcast from hours to under 30 seconds.
  • Universal Accessibility: Completely eliminated the literacy and language barrier by prioritizing native-language voice broadcasts.
  • Civic-Grade Human-in-the-Loop UX: Designed an intuitive dashboard giving city officers one-click authorization over AI suggestions, preventing accidental public panic.

What we learned

  • Civic Tech is an Equity Challenge: Complex AI is useless if advice cannot be understood by a daily-wage earner who speaks only a local dialect—voice-first design is essential.
  • Grounded Prompts Win: Feeding Gemini infrastructure context (such as the number of schools nearby) transforms generic warnings into tactical, life-saving municipal playbooks.

What's next for AeroHealth AI

  • Predictive Forecasting: Incorporating wind vector and microclimate models to predict ward pollution spikes 6 to 12 hours before they occur.
  • Interactive IVR & WhatsApp Dispatch: Automated outbound voice calls (IVR) and WhatsApp Business bot integration to reach basic feature phones without app installs.
  • Hospital Surge Prediction: Linking ward vulnerability scores directly to municipal hospital ER systems to forecast asthma and cardiac admission surges.
Share this project:

Updates

Submission history