EquityMap | Devpost

Inspiration

On July 4, 2025, Congress passed the "One Big Beautiful Bill," a sweeping federal budget reconciliation act that introduced major changes to social safety net programs. Among them: an 80-hour/month work requirement for SNAP benefits for adults 18–59, tightened Medicaid eligibility thresholds, and reduced funding for community health programs, changes projected to impact over 5 million Americans. In Washington, D.C., residents in underserved communities now face "Benefit Deserts" where food assistance and healthcare are difficult to locate and even harder to understand. We built EquityMap to close that gap.

What it does

EquityMap is an AI-powered geospatial platform that helps D.C. residents find food assistance and healthcare. It maps 450+ verified SNAP retailers and primary care facilities using real government data, lets users search by address to find the nearest resources, and includes an AI assistant (EquityGuide) that explains benefit eligibility and walks users through a SNAP screener in under 30 seconds. The entire platform supports 6 languages and works in dark mode.

How we built it

We built the frontend with React, TypeScript, and Tailwind CSS. The interactive map uses Leaflet.js with CartoDB tiles, loading real datasets (USDA SNAP retailers and DC Primary Care Facilities) client-side with PapaParse. The AI chatbot uses Google's Generative AI SDK to call Gemini 2.5 Flash with a custom system instruction that adapts to the user's selected language. Address search and validation use OpenStreetMap's Nominatim geocoding API. Community resource submissions go through geocoded address validation and CAPTCHA verification. We deployed on Vercel.

Challenges we ran into

Gemini API quota limits were our biggest obstacle — we burned through the free tier mid-development and had to switch models multiple times (gemini-2.0-flash → gemma-3-1b-it → gemini-2.5-flash) before enabling billing resolved it. The health facility CSV used Web Mercator coordinates instead of lat/lng, requiring a projection conversion. We also had to geofence address validation to D.C. boundaries because Nominatim would return results for almost anything.

Accomplishments that we're proud of

The eligibility screener is entirely deterministic — it works instantly without calling the AI, so it's reliable even if the API goes down. The multi-language support translates both the UI and AI responses across 6 languages with a single toggle. Every resource on the map is backed by real, verified government data — nothing is fabricated. The step-up verification on community submissions uses address geofencing and CAPTCHA to prevent abuse.

What we learned

Working with multiple AI models taught us that not all models support the same features (Gemma doesn't support system instructions, for example). We learned the importance of fallback mechanisms — having curated responses ready when the AI is unavailable kept the app functional during quota issues. We also gained experience with geospatial data processing, coordinate system conversions, and building multilingual interfaces.

What's next for EquityMap

Expanding beyond D.C. to other cities with similar benefit deserts. Adding a backend to persist community-contributed resources with admin review. Integrating real-time benefit application status tracking. Building a mobile app for offline access in areas with limited connectivity.