Google Solutions Challenge Submission
Turning paper field reports into real-time community intelligence.
🔗 Live Demo: uplift-c2f35.web.app
Local NGOs and social groups collect critical community data through handwritten paper surveys and field reports. This valuable intelligence gets trapped — scattered across offices, clipboards, and WhatsApp groups — making it impossible to see the biggest problems clearly or coordinate volunteers efficiently.
UpLift is a two-role platform that bridges the gap between field data and coordinated action:
- NGO Staff upload photos of handwritten reports → AI extracts structured data → human verifies → report appears on a live map
- Volunteers chat with an AI coordinator → AI queries nearby verified tasks → volunteers get matched to the most urgent needs
| 
|
| 
Upload a photo of any messy handwritten field report. Gemini 1.5 Flash extracts text, categorizes the issue, assigns an urgency score (1–5) using a fixed rubric, and geocodes the location — all in one pass.
Every AI-processed report lands in a Verification Queue before going live. NGO staff can view the original image side-by-side with the AI's extracted text (the Insight Trace), edit any field, then Approve or Reject. The AI assists; humans decide.
Volunteers describe their location and skills in plain language. Gemini uses Function Calling to search verified tasks via findNearbyTasks() — it cannot make up results or drift off-topic. Matched tasks appear in a side panel with a mini-map.
Full English ↔ Hindi translation across every UI string. Toggle in one click from the header.
A floating SOS button detects the user's location via browser geolocation and submits a Critical (5/5 urgency) alert instantly.
NGO coordinators can export all reports to a spreadsheet with one click for offline analysis or donor reporting.
User uploads photo
↓
React (Vite) frontend
↓
Gemini 1.5 Flash API (client-side, JSON mode)
├── OCR + text extraction
├── Urgency scoring (rubric-anchored)
├── Category classification
└── Location extraction
↓
Nominatim geocoding (OpenStreetMap, free)
↓
Firebase Firestore (status: "pending")
↓
NGO Verification Queue → Approve
↓
status: "verified" → Live map
↓
Volunteer Chat → Gemini Function Calling → findNearbyTasks() → Firestore
| Layer | Technology | Cost |
|---|---|---|
| Frontend | React 19 + Vite 8 | Free |
| Maps | Leaflet.js + OpenStreetMap | Free |
| Geocoding | Nominatim API | Free |
| Database | Firebase Firestore (Spark) | Free |
| Hosting | Firebase Hosting | Free |
| AI | Gemini 1.5 Flash (Google AI Studio) | Free tier |
| Icons | lucide-react | Free |
Total infrastructure cost: $0
| Principle | Implementation |
|---|---|
| Human-in-the-Loop | No report reaches the live map without NGO staff approval |
| Transparency | Insight Trace shows the exact text the AI analyzed |
| Constrained Agent | Volunteer chat AI can only call findNearbyTasks() — cannot modify data or answer off-topic queries |
| Graceful Failure | Every failure path writes an errorMessage to Firestore and shows a Retry button — no data is silently lost |
| Consistent Scoring | Few-shot rubric anchoring prevents urgency score drift across reports |
| Role | How to Access |
|---|---|
| Volunteer | Any phone number + any 6-digit OTP + any name |
| NGO Staff | Any phone + any OTP + any name + org code: UPLIFT2026 |
Demo Zone: 25 pre-seeded reports in Indiranagar, Bangalore — spread across 3 weeks of realistic timestamps, covering all urgency levels and categories.
- Node.js v20+
- npm
# Clone the repo
git clone https://github.com/YOUR_USERNAME/uplift.git
cd uplift
# Install dependencies
npm install
# Configure your keys (see below)
# Then start the dev server
npm run devApp runs at http://localhost:5173
| Command | Description |
|---|---|
npm run dev |
Start development server with hot reload |
npm run build |
Build for production (output: dist/) |
npm run preview |
Preview production build locally |
src/
├── config/
│ ├── firebase.js # Firestore connection (gitignored)
│ └── gemini.js # Gemini API models (gitignored)
├── components/
│ ├── RoleSelection.jsx # Landing page with animated stats
│ ├── LoginPage.jsx # Phone + OTP + NGO verification
│ ├── Header.jsx # Role indicator, language toggle, upload
│ ├── MapView.jsx # Leaflet map with urgency-coded markers
│ ├── ReportCard.jsx # Verification queue card with actions
│ ├── InsightModal.jsx # AI Insight Trace overlay
│ ├── UploadModal.jsx # Drag-drop upload with progress stepper
│ ├── SOSButton.jsx # Floating emergency alert button
│ ├── StatusBadge.jsx # Urgency + status badge components
│ └── Toast.jsx # Toast notification system
├── pages/
│ ├── NGODashboard.jsx # Map + Verification Queue layout
│ └── VolunteerChat.jsx # Chat + Matched Tasks + mini-map
├── hooks/
│ ├── useReports.js # Firestore real-time listener (falls back to seed data)
│ ├── useUpload.js # ID-anchored upload + Gemini analysis pipeline
│ └── useChat.js # Gemini function calling chat agent
├── i18n/
│ ├── LanguageContext.jsx # React context for EN/HI switching
│ └── translations.js # Complete EN + HI translation strings
└── utils/
├── demoZone.js # 25 seed reports + geospatial utilities
├── geocoding.js # Nominatim + local fallback lookup table
└── exportCSV.js # Report export with Unicode BOM for Excel
A key technical decision that prevents the most common live-demo failure (frozen progress bars):
// 1. Frontend generates the document ID BEFORE upload begins
const reportId = doc(collection(db, 'reports')).id;
// 2. Image is uploaded to Storage using that same ID as filename
// 3. Firestore listener is attached to reports/{reportId} immediately
// 4. All status updates from the AI pipeline write to that same doc
// Result: the UI is always listening to the exact right document
// from the moment the user clicks uploadGoogle Solutions Challenge — Problem Statement: "Design a powerful system that gathers scattered community information to clearly show the most urgent local needs. Build a smart way to quickly match and connect available volunteers with specific tasks and areas where they are needed most."
 Samarth230 |
 rohanrpais |
MIT
