Debris AI is a hackathon project that analyzes images of construction debris to identify materials, estimate reuse feasibility, and provide safe, non-structural reuse suggestions.
The project is organized into the following directories:
frontend/: Contains the user interface for the application.backend/: Contains the core backend logic, including the API server, Vertex AI integration, and Kafka streaming.models/: Contains a conceptual notebook for training a verifier model.deployment/: Contains documentation for setting up the project on Google Cloud and Confluent Cloud.The backend is a Python Flask application designed to be deployed on Cloud Run. It exposes two endpoints:
POST /analyze-image: Accepts an image of construction debris, analyzes it using Gemini, and returns a JSON response with material identification, reuse feasibility, and reuse suggestions.POST /chat: Accepts a text query and returns a response from Gemini.The backend also integrates with Confluent Cloud to stream events for every request.
Debris AI prioritizes safety by providing a disclaimer with every analysis and offering conservative, non-structural reuse suggestions. A verification layer is also in place to re-prompt the model with stricter instructions if the initial confidence score is low.
To get started with the project, please refer to the documentation in the deployment/ directory.