Inspiration

Driven by the urgent need to combat rising pollution and habitat loss, our team created an AI solution to bridge the gap between human habits and effective waste management. Today's global footprint is plagued by poor sorting practices, centuries-long plastic persistence, and catastrophic food waste. Our platform harnesses computer vision to instantly identify waste streams, calculate material decomposition lifetimes, and suggest actionable upcycling pathways. Furthermore, it transforms near-expiry food items into custom recipes—preventing unnecessary waste before it happens. We are using modern AI to make waste sorting effortless and bring practical sustainability into every home.

What it does

Relast is an AI sustainability assistant for your phone that helps you get more life out of what you already own. Snap a photo or type a question, and it tells you what to do with the item:

Sort it right: which bin it belongs in (recycling, garbage, compost, hazardous or e-waste), plus ways to reuse it.
Give it a second life: how long a product typically lasts, repairs that would extend it, and where to donate or resell it.
Rescue your leftovers: quick recipes that use up the food you have, and tips for storing it longer.

Google's Gemini writes each answer. For waste sorting, a garbage-classification model runs right on the phone to pick the bin, and Gemini explains why.

Relast also estimates the CO₂ you save by walking instead of driving, based on how fast you're moving while the app is open. Profile shows a weekly chart and Home shows today's total.

How we built it

App

Built with Expo and React Native. Expo Router handles navigation between Home, the assistant, answers and Profile.
The assistant screen's spinning category wheel is animated with Reanimated and Gesture Handler. Users can take a photo with the camera, choose one from their library, or just type. #### AI
Gemini 3.5 Flash-Lite reads the photo and text. We gave it per-category instructions and required a fixed JSON format for every reply, so each answer comes back in a predictable shape: the item, the advice, recipes or repair options, and a confidence level the screens can show.
For waste sorting we added a Vision Transformer trained on 12 garbage types. We converted it to ONNX and compressed it to 8-bit (about 83 MB) so it runs on the phone with Transformers.js. When it's confident, its pick is passed to Gemini, which explains the answer.
If the on-device model isn't available, the app can fall back to a small Python FastAPI server running the same model. #### Backend
Supabase handles accounts, stores each user's question history and trips in Postgres, and keeps uploaded photos in private storage. Row-level security means users can only see their own data.
Database changes are versioned as migrations and applied with the Supabase CLI. #### Carbon tracking
The phone's location service reports how fast you're moving while the app is open. Speed puts each stretch of travel into walking, car/bus, train or flight.
$Distance × average$ emissions per km gives an estimate of CO₂ used and CO₂ saved by walking. #### Process
We split the work across feature branches with pull requests (UI, AI, history, sorting, carbon tracking), and used AI coding assistants to move faster during the hackathon.

Challenges we ran into

Since Relast is designed for on-the-go waste scanning, ensuring seamless mobile web access was critical. However, we hit a major roadblock when mobile browsers repeatedly crashed upon requesting device camera permissions. To resolve this, we systematically debugged our camera API calls and cross-tested each route across multiple mobile viewports. Through targeted stress-testing, we isolated the underlying web API bug and successfully enabled reliable, real-time camera scanning on mobile web.

Accomplishments that we're proud of

Stepping into Hack The Valley as complete hackathon beginners, building a functional, end-to-end product in 36 hours was a major milestone for us. Beyond overcoming tight time constraints and rapid technical debugging, we are immensely proud to have turned an ambitious concept into a tangible, high-impact tool for sustainability. Knowing that Relast directly addresses global challenges like waste sorting, food waste, and environmental education makes this achievement incredibly rewarding. We are proud of our resilience, our rapid learning curve, and our shared vision to build technology that leaves a lasting positive mark on the planet.

What we learned

We learned how challenging and stressful hackathons can be. Working under pressure, racing against a tight deadline, and collaborating with a group of people we had just met pushed us outside our comfort zones and taught us how to adapt to high-pressure situations.

We also learned the importance of teamwork. Everyone brought unique skills, perspectives, and ideas to the table, and it was through combining our individual contributions that we were able to bring our idea to life.

Finally, we learned the value of iterating on ideas. We started with an initial vision for our project, but by discussing different approaches, challenging our assumptions, and critically evaluating our ideas, we were able to develop an even stronger plan. This experience taught us that great ideas don't always come together on the first try; they improve through collaboration, reflection, and a willingness to adapt.

What's next for Relast

Beyond Hack The Valley, our goal is to scale Relast into a feature-rich, community-driven platform that makes sustainable habits addictive and actionable. Our immediate roadmap includes:

Gamified Leaderboards & Community Impact: Introduce competitive local and global leaderboards to track real-time carbon offsets, driving friendly competition among schools, neighborhoods, and organizations.
Dynamic Achievements & Badges: Reward users with milestone badges (e.g., "Zero-Waste Chef" or "100kg Plastic Saver") to boost long-term retention and validate eco-friendly choices.
Social & Peer-to-Peer Food Donation Hub: Build a neighborhood connection feature where users can list surplus ingredients or cooked food for direct local pickup, turning potential waste into immediate community support.
Enhanced AI Precision & UX: Expand our multi-modal vision models to identify complex composite materials and deliver even faster low-latency camera scans on any mobile web browser.