Inspiration🔥 Inspiration
Our team was inspired by the overflowing waste bins in school cafeterias — where compostable materials are often tossed in trash and recyclables go to waste. We realized that most people want to do the right thing, but they’re confused by vague packaging or unclear signage. We set out to build an AI-powered assistant that could help students instantly classify their waste and reduce landfill impact.
🧠 What it does FlaminGo is a real-time waste classification tool that:
Uses computer vision to detect whether an item is biodegradable or not.
Extracts text labels from the object using OCR to boost prediction confidence.
Leverages GPT-4o to explain the classification and offer insights.
Offers both webcam-based and image upload classification.
Stores and displays recent classifications, and allows users to query the assistant about trends or errors. 🛠️ How we built it Frontend: Tailwind CSS, HTML, and vanilla JS for a clean, mobile-friendly UI.
Backend: Flask handles image capture, classification, and integration with MongoDB for login/auth.
Computer Vision: Trained a CNN model (based on VGG16) to classify waste items into 5 categories.
OCR: Used pytesseract to extract any visible text on the object.
GPT-4o: Sends recent results (including OCR and images) to OpenAI to interpret user activity.
Authentication: Built a secure login/register system with hashed passwords and MongoDB for user sessions.
🧩 Challenges we ran into Managing realtime webcam input and converting it into usable model input.
Handling image hosting vs base64 conversion to support GPT-4o vision input.
Cleaning up the dataset to ensure meaningful biodegradable vs non-biodegradable classification.
Getting OpenAI’s new API working smoothly with Flask and async vision uploads.
Ensuring performance was snappy even with model inference and OCR combined.
🏆 Accomplishments that we're proud of Created a fully functional AI-powered app with vision + language + interactivity in under 48 hours.
Integrated OpenAI’s GPT-4o vision API to deliver true assistant-style interpretation.
Designed an interface that’s both fun and functional for school environments.
Built a secure login system with password hashing and database support.
Enabled intelligent conversations between students and the assistant around sustainability. 📚 What we learned How to deploy vision models in a real-world app.
Deepened our understanding of OCR limitations and how GPT can help fill in the gaps.
Learned the nuances of OpenAI's latest vision API and how to format prompts effectively.
Gained hands-on experience with frontend/backend integration using Flask + Tailwind.
Realized the importance of user feedback loops in making AI feel helpful and not confusing.
🚀 What's next for FlaminGo Add multi-object detection to support trays with multiple items.
Use YOLOv8 for smarter bounding boxes and class prediction.
Provide personal stats and insights on how much waste each user diverted.
Explore classroom-wide dashboards to gamify sustainable behavior.
Deploy the tool on a school tablet kiosk or even as a browser extension near trash bins.
Log in or sign up for Devpost to join the conversation.