SaansCare

SaansCare turns Lahore's air-quality data into a live Gov + Resident dashboard - predicting 12-month health risk, tracking vehicle emissions to the owner, and mapping pollution hotspots city-wide.

Comment

Inspiration

Lahore regularly ranks among the world's most polluted cities, especially during winter smog season (Nov–Feb), when AQI routinely crosses "Hazardous." Public AQI numbers exist, but they're just a number — they don't tell a parent whether it's safe for their asthmatic child to play outside today, and they don't give the Health Department a way to see where the pollution-driven health burden is actually concentrated. The Smart City Hackathon Lahore 2026's problem statement — "Linking Air Pollution to Public Health Outcomes," for Punjab EPA and the Health Department — was exactly the gap we wanted to close.

What it does

SaansCare is a role-based dashboard, not just a data viewer:

  • Residents get their district's current AQI, a 2-year trend, an AI-generated 12-month health risk outlook, and can register their own vehicle(s) for emission/maintenance tracking.
  • Gov/EPA officials get city-wide risk exposure by district, a live device and Safe City camera network, road-segment pollution tracking, and a vehicle registry auto-flagged for overdue maintenance — traceable to the owner.
  • Administrators manage the whole platform: provisioning Gov accounts directly (no open registration for that role), editing or removing accounts, and running scoped data resets.

How we built it

MERN-style stack, but with MySQL instead of Mongo: Node/Express/Sequelize on the backend, React 19 + Vite + Tailwind on the frontend, Chart.js and Leaflet for visualization, and Groq's LLM for the narrative forecasts (with a deterministic rule-based fallback so the feature never breaks live if the API is unavailable). We seeded ~5,800 historical AQI readings across 8 real Lahore districts with realistic winter-smog seasonality, so every chart and forecast works against real patterns, not flat placeholder data.

We built it iteratively: core AQI dashboard first, then the Gov/User dual portal, then real MySQL (swapped in from an initial SQLite prototype), then the admin layer, PDF reporting, and UI polish — testing each layer end-to-end against a live database before moving to the next.

Challenges we ran into

  • A model deprecation, not a bug. Our AI forecasts kept silently falling back to the rule-based generator even with a valid API key — turned out Groq had deprecated the model we were using. Tracked it down properly instead of guessing, and fixed the actual cause.
  • Being honest about what's real. Lahore's actual Safe City camera network is a closed government system with no public API. Rather than fake a live feed, we built a clearly-labeled simulated view with a link to genuine public reference footage — useful for the demo without misrepresenting what's actually connected.
  • A subtle data bug. Road tracking logs multiple historical readings per road for trend purposes, but the API was returning all of them unaggregated — 112 near-duplicate cards instead of 8 clean ones. Fixed at the source with proper aggregation.
  • Confidentiality by design, not as an afterthought — locking registration to residents only, making Gov/Admin provisioning deliberate and traceable, and validating CNIC/email formats on both client and server.

What we learned

Building for three genuinely different personas (resident, official, administrator) in one system forces real product decisions, not just more UI — what each role should see, edit, and never be able to touch. We also learned to treat "AI-powered" features defensively: always design the fallback path first, then layer the LLM on top, so a third-party API hiccup never takes down the core product.

What's next

Swapping the seeded historical data for a live ingestion job against Punjab EPA's public AQI feed, and pursuing real Safe City camera access through proper channels rather than simulating it.

Built With

Share this project:

Updates

Submission history