Inspiration

Climate change is one of the most pressing challenges of our time, yet many people don't have visibility into their daily carbon emissions. I wanted to make carbon tracking effortless and autonomous through this project

What it does

Re-Vision is an AI-powered carbon footprint tracker that uses your device's camera to capture and analyze daily activities in real-time. Users start a continuous capture session that intelligently collects 12-frame snapshots, processes them with Google's Gemini 2.0 Flash model to identify transportation and consumption activities, and calculates cumulative carbon emissions. The app includes voice-based chat with ElevenLabs TTS for natural conversational insights, a leaderboard to compare progress with friends, and a 7-day trends dashboard with detailed analytics.

How I built it

I built Re-Vision using Next.js 16 for the frontend and API layer, MongoDB Atlas for the database, and Google Gemini 2.0 Flash for activity recognition. I used perceptual hashing to identify similar frames similar frames and discard them. I integrated ElevenLabs TTS for conversational feedback and I used the browser for voice recognition

Challenges I ran into

The biggest challenge was handling continuous video capture at 30fps—generating an API call for every frame would instantly hit rate limits and incur massive costs. I solved this by implementing perceptual hashing to compare frame similarity and only process frames that differ sufficiently (hamming distance ≥ 23), reducing API calls by ~70%. I also had to manage asynchronous analysis queues with a max of 3 pending jobs to prevent overwhelming the backend while keeping the UI responsive.

Accomplishments that I am proud of

I'm proud of building a seamless, production-ready camera capture pipeline with intelligent frame deduplication. The perceptual hashing implementation eliminated redundant API calls while maintaining high accuracy. The multi-modal interface—combining real-time camera capture, AI analysis, voice chat, and data visualization—creates a truly engaging user experience that makes carbon tracking feel interactive and rewarding.

What I learned

I learned how to implement perceptual hashing for efficient image deduplication in real-time systems and gained hands-on experience with MongoDB vector search capabilities.

What's next for Re-Vision

I plan to move this application onto the ray ban meta glasses. In addition I want to integrate this project with Apple Health and Google Fit for seamless activity tracking, and add gamification features like challenges and team competitions.

Built With

Share this project:

Updates