IndoHood
IndoHood turns domestic waste into economic value using Amazon Bedrock (Nova Pro) for multimodal waste classification and live mandi rate estimation. Built serverless on AWS Lambda, DynamoDB, and Amazon Location Service, it automates doorstep pickups and gamifies civic segregation with neighborhood rankings.
IndoHood: Making Waste Segregation Rewarding
Team X (X = finding real value) · Ramanujan College, University of Delhi · Environmental Hacks, Waste and Energy track
The problem
I have segregated my waste at home for a long time. Then I saw the picker dump everything in one place, and all my effort mixed back together. That hurt.
Around me, I noticed two things. People know very little about what goes where. And even those who know often don't bother, because they get nothing in return, or they say they have no time. Throwing mixed waste is just easier.
The problem isn't that people don't care. Nobody guides them, nobody rewards them, and the system after pickup doesn't respect their effort.
What we built
IndoHood (Indo for India, hood for our surroundings) is a web app. We first called it "Dobara", then renamed it to fit our Indian context of reuse and sustainable living.
A resident points the camera at a waste item and gets the category, segregation advice based on Indian waste rules, and an estimated weight and scrap value from mandi rates. They schedule a pickup, a local collector completes it, and the resident earns EcoCredits, a leaderboard rank and a Green Citizenship Certificate. There are two portals: one for residents and one for pickers.
Rank near you. The leaderboard first shows a user's rank among all IndoHood users, then their rank within their local area. When the user's device sends its GPS coordinates (the browser asks for permission first), our Lambda calls Amazon Location Service (Place Index) to do reverse geocoding: it turns the numbers into a real place name, such as the neighbourhood, city and postal code. We group users by postal code, so people living near each other are compared together. We store only the area name, not the coordinates. The score stays the same, only the comparison group changes when the user moves to a new area. The ranking logic itself is our own code.
How it runs on AWS
• AWS Amplify: hosts the React + Vite frontend with CI/CD
• API Gateway (HTTP API) + Lambda: our Express backend on Node.js 20, wrapped with serverless-http
• Amazon Bedrock (Amazon Nova Pro): reads the photo and returns structured JSON for waste classification
• Amazon Location Service (Place Index): reverse geocoding for the local rank
• DynamoDB: pickups and users, synced across devices Everything runs in one region (ap-southeast-2) to keep requests short.
Our four days
Day 1 (Oct 8): We spent most of the day on the idea, the problem, the solution, the full user flow and the AWS architecture. Ridhi designed the frontend while we discussed architecture, and we started building by evening.
Day 2 (Oct 9): The frontend was nearly done by midday. I then gave the AI the context it needed: live scrap prices, Indian waste management rules and home reuse ideas, which Kanishka researched. I built the backend: ranking, CO prevention calculation and impact tabs. Around midnight I added the
2
database, since Day 3 would go to the DTU build day.
Day 3 (Oct 10): At DTU we tested the whole app and found things still broken: the ranking, stock images in the activity feed instead of the scanned photo, real login, and data not being stored. We fixed them there, showed the project to other builders, and the feedback was positive.
Day 4 (Oct 11): We added the location-based rank using Amazon Location Service, then did final fixes, the video and the submission.
What fought back
• The AI didn't detect items or set categories properly at first. Early scans were wrong two or three times. We fixed it by giving the model our own conditions: Indian waste rules, reference ranges and scrap prices.
• New users got a rank. Accounts with zero activity showed up as "#3". Now they show as unranked until they complete a pickup.
• One user's activity showed on other accounts, and later one user's details showed for everyone. We fixed the data isolation so each user sees only their own activity.
• The database wasn't saving data. Fixing it exposed the isolation bug above, which only appeared with real data.
• Stock images instead of real photos. Activity cards now show the photo the user actually scanned.
Most of these only appeared when we tested with real data and real people, not with pre-stored data.
Testing
We scanned 25+ real household items (kitchen scraps, cardboard, PET bottles, cables and more). Early tests didn't give the results we wanted. After our fixes it worked well, and at the DTU build day real users tried it and gave positive feedback.
Tools and AI we used
• Antigravity for coding
• Amazon Nova Pro on Bedrock as the AI inside the platform
• ChatGPT for research
What's next (roadmap, not built yet)
• IndoHood Smart Bin (our big next step): a machine placed in the city that uses the same AI to recognise an item, drops it into the right bin (Dry, Wet, E-Waste/Hazardous), credits the resident, and alerts pickers when it is full. We have not built the hardware yet. This weekend we built the software side it will run on: AI scan, pickups, credits and DynamoDB.
• Route optimization for pickers: group nearby pickups into shorter routes
• WhatsApp booking: send a photo on WhatsApp to book a pickup, for people who don't use web apps
• Scaling local rank: index users by postal code so rankings stay fast with many users • Corporate CSR offsets are a later idea
Credits
Mandi rates: ScrapMantra, with approximate Delhi-NCR benchmark rates as fallback. Libraries:
React, Vite, Tailwind CSS, lucide-react, canvas-confetti, Express, serverless-http, AWS SDK v3.
Fonts: Inter, Playfair Display.
Team X: Aakash (lead, engineering), Ridhi Kumari (UI/UX, video), Kanishka Verma (research)
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article