Envolve | Devpost

Landing Page
Local Issues
Object Detection
Points Leaderboard
System Overview

Inspiration

Imagine your neighborhood. Small environmental issues—litter, illegal dumping, mismanaged waste—often go unnoticed or unaddressed. Yet these minor problems can quickly snowball, deteriorating local quality of life and contributing to broader issues like urban decay and pollution. Research shows that even a slight uptick in household waste mismanagement can trigger significant environmental damage. For instance, the World Bank’s What a Waste 2.0 report projects global waste to surge from 2.01 billion tonnes in 2016 to 3.40 billion tonnes by 2050, with at least 33% mismanaged.

What it does

Our solution is simple: a community-driven platform that empowers residents to capture and share these issues with just a click. Snap a photo, add a quick note, and let our AI do the heavy lifting—identifying critical problems, categorizing them, and even initiating polls that automatically connect you with local authorities.

Intelligent Reporting & Analysis:

Residents capture a photo of a local issue with a title, and the platform auto-captures location data while an AI model recognizes the object. The combined details are then processed by a generative model that suggests a detailed description, relevant tags, and a severity assessment to prioritize action.

Real-Time Mapping & Automation:

Reported issues are instantly aggregated on an interactive map, highlighting neighborhood hotspots. When community polls indicate critical concern, the system automatically connects with the appropriate authorities and initiates a call on behalf of the users.

How we built it

Frontend: Built with Next.js/React and TailwindCSS for a smooth and responsive user experience.

Backend: Powered by FastAPI and NextServer, handling requests efficiently.

Database: Supabase for effecient querying, and scaling.

AI / ML: Used a transfer-learned YOLO v11 model fine-tuned on a 40k-image dataset (including light posts, floods, potholes, and litter) using Roboflow, with OpenCV on Modal for real-time image processing. Integrations with FastAPI, Hume, Twilio, and Anthropic API drive intelligent analysis and automated workflows.

Challenges we ran into

We had to first gather the datasets for obscure things like street lights and flooding, then we had a misunderstanding for transfer learning, where we thought we would still have the existing classes and my classes on top, which was incorrect, so we ran into the trouble of trying to keep the pre-trained classes and the classes we've trained: https://y-t-g.github.io/tutorials/yolov8n-add-classes/. We had to change ultralytics underlying code to be able to freeze a certain number of layers and be able to train mine on top

Accomplishments that we're proud of

Intuitive and easy to use interface
Multiple ways to facilitate action amongst a community of people to bring attention to a environmental issue, from automating call to 311 to auto completing categories, tags and more
Understanding and learning superbase subscriptions