UrbanOre - AI-Powered E-Waste Management Platform

Home Page

Inspiration

E-waste is one of the fastest-growing waste streams globally, with over 50 million tonnes generated annually. In India alone, informal recyclers ("Kabadiwalas") handle most e-waste, often without proper safety knowledge or fair pricing. We saw three critical problems:

Unsafe handling — Workers unknowingly expose themselves to hazardous materials like lithium batteries and mercury
Data security gaps — Discarded devices contain sensitive personal data that rarely gets properly wiped
Value loss — Sellers don't know the true worth of their devices, while buyers lack trust in listings

We were inspired to bridge the gap between technology and sustainability — using AI to make e-waste recycling safer, more transparent, and more profitable for everyone.

What it does

UrbanOre is an AI-powered e-waste management platform that transforms how electronic waste is identified, certified, and traded:

AI Device Analysis — Upload a photo and Gemini 1.5 Pro Vision instantly identifies device type, brand, model, and condition
Safety-First Hazard Detection — Real-time warnings for dangerous components like swollen batteries, capacitors, and toxic materials
Data Wipe Verification — Secure data destruction verification with downloadable PDF certificates for laptops and PCs
Smart Value Estimation — Calculates both reuse and scrap value with detailed metal breakdown (gold, copper, palladium in circuit boards)
Live Auctions — Connects e-waste sellers directly with verified recyclers through real-time WebSocket-powered bidding
Environmental Impact Tracking — Shows CO₂ savings and sustainability metrics for every transaction

How we built it

We designed UrbanOre as a microservices architecture to ensure scalability and maintainability:

Frontend

React 18 with Vite for a fast, responsive single-page application
CSS Modules for component-scoped styling with a clean, professional white theme
React Router for seamless navigation between analyze, auction, and certificate pages

Backend Services

Express.js server integrating with Google Gemini 1.5 Pro API for AI vision analysis
WebSocket (ws) implementation for real-time auction bidding
PDFKit for generating tamper-evident data destruction certificates
JWT + bcrypt for secure user authentication

Infrastructure

Redis for message brokering, caching, and pub/sub between microservices
Docker Compose orchestrating 5 independent services:
- Backend Service (API gateway)
- Seller Service (AI analysis + listings)
- Buyer Service (bidding + watchlists)
- Auction Engine (bid matching + anti-sniping)
- Notification Service (alerts)

AI Integration

We crafted detailed system prompts for Gemini to return structured JSON responses covering device identification, component analysis, safety hazards, and value estimation — all from a single image upload.

Challenges we ran into

Structuring AI Responses — Getting Gemini to consistently return valid, parseable JSON with all required fields required extensive prompt engineering and validation layers
Real-time Auction Synchronization — Implementing WebSocket connections that stay in sync across multiple bidders while preventing race conditions in bid placement
Hazard Detection Accuracy — Training the AI to reliably identify dangerous components (swollen batteries, leaking capacitors) from various image angles and lighting conditions
Data Wipe Verification — Designing a trustworthy verification flow that provides genuine security assurance without access to the actual device storage
Microservices Communication — Coordinating state between 5 independent services through Redis pub/sub while maintaining data consistency

Accomplishments that we're proud of

End-to-end AI workflow — From photo upload to certified PDF certificate in under 30 seconds
Production-ready architecture — Fully containerized microservices that can scale independently
Safety-first design — Prominent, color-coded hazard warnings that could genuinely prevent injuries
Real market value analysis — Detailed breakdown of precious metals (gold, palladium, copper) in circuit boards
Live auction system — Functional WebSocket-based bidding with anti-sniping protection
Beautiful, accessible UI — Clean professional design that works across devices

What we learned

Prompt Engineering is crucial — The quality of AI responses depends heavily on how you structure system prompts and validation
Microservices add complexity — While scalable, coordinating multiple services requires careful planning of communication patterns
Real-time systems are hard — WebSocket state management and race condition handling taught us a lot about distributed systems
E-waste is a massive problem — Researching the industry revealed how much valuable material is lost and how many workers are at risk
AI vision has limitations — Gemini performs differently based on image quality, angle, and lighting — requiring robust error handling

What's next for UrbanOre

Mobile App — React Native app with camera integration for on-the-spot device scanning
AR Disassembly Guides — Augmented reality overlays showing safe step-by-step device teardown procedures
Blockchain Certificates — Immutable data destruction certificates stored on-chain for enterprise compliance
Kabadiwala Verification — KYC and rating system for recyclers to build trust in the marketplace
Pickup Scheduling — Integrated logistics for doorstep e-waste collection
Government Integration — Partnerships with municipal e-waste programs and EPR (Extended Producer Responsibility) compliance
Multi-language Support — Hindi, Tamil, and regional language interfaces for broader accessibility