Welcome to Food Decoder!
Food Decoder is an open source, lightweight full stack web application designed to help people know what they are consuming on daily basis. Helps maintain dieting, body building plans and helps in maintaining a healthy lifestyle using AI models and Ingredients(fetched from world.openfoodfacts.org) in the packaged food.
Features
-
Live Barcode Scanning: Native in-browser scanning using device cameras (no app download required).
-
Vision AI Fallback: Upload a photo of a food label if the barcode isn't recognized, utilizing Vision LLMs to extract ingredient text automatically. (This uses Nvidia nemotron nano 12b v2 vl)
-
AI Health Analysis: Automatically generates a Health Score , highlights positive nutritional traits, and flags potential concerns (like hidden sugars, artificial additives, emulsifiers, food dyes etc.).
-
User Friendly UI: A premium, responsive frontend featuring glassmorphism, 3D interactive cards, magnetic buttons, and smooth momentum scrolling.
-
Resilient Backend: A FastAPI backend hosted on Hugging Face that seamlessly routes traffic between Gemini 2.5 Flash and an OpenRouter fallback (Nemotron) to ensure 100% uptime.
Tech Stack
Frontend
-
HTML5, CSS3, Vanilla JavaScript
-
HTML5-QRCode (For client-side barcode scanning)
-
Custom CSS animations & 3D tilt effects(3d effects are only visible in desktop version)
Backend & AI
-
Python Framework: Fastapi
-
Hosting: Hugging Face Spaces
-
Primary LLM: Google Gemini 2.5 Flash
-
Fallback LLM: Nvidia Nemotron-12b (via OpenRouter)
-
Food Database: Open Food Facts API
How does it work??
-
Scan or Upload: If the database contains the product scanning the barcode(usually behind the food package) can give instant insights but if you want deeper analysis or the product is not in the database you can take a photo of the backside of the product which will give a more detailed insight.
-
Database Lookup: If a barcode is scanned, the app pings the Open Food Facts API to retrieve the raw ingredient string.
-
AI Processing: The text (or image) is sent securely to the FastAPI backend. A highly specific prompt forces the LLM to act as a toxicologist/nutritionist, returning a strictly formatted JSON object.
-
Interactive Report: The frontend parses the JSON and renders a beautiful, color-coded report card detailing the food's safety profile.
Local Installation
Idk why would anyone install it locally but here are the steps to do so:-
- Clone the Repo
git clone https://github.com/adityaprasad-sudo/FoodAnalayser.git
cd FoodAnalayser
- Setup the api keys
- make .env file in the cloned repo
this goes into your .env file and put your open ai api keys(but since i am using hackclub api's the client url would be hackclub so make sure you change that to your needs)
`OPENAI_API_KEY = "Your api key"`
3.Install dependencies
pip install fastapi uvicorn openai pillow requests python-dotenv python-multipart
uvicorn
4.Runing the frontend
- You can serve the HTML/CSS/JS files using any standard live server (like the VS Code Live Server extension)
Note: Camera access in browsers requires either a localhost or https:// connection
Disclaimer
DecodeFood is built for educational and informational purposes only. The AI-generated health scores and insights do not constitute professional medical advice. Users with severe food allergies should always verify ingredients on the physical packaging.
Author
Built with ❤️ by Aditya Prasad * GitHub: @adityaprasad-sudo
Flowchart
Screenshots and Demo
item used in the video - Amazon
- Demo Video
(https://drive.google.com/file/d/1url-S7SR59TI-1AMfo4ylVShVGI_h3xj/view?usp=sharing)
- Demo Screenshots
Which Barcode to Scan?
