NutriScan — AI-Powered Indian Food Nutrition Bot

NutriScan is an end-to-end AI system that helps users understand the nutritional value of Indian meals by simply uploading a food image on Telegram.

Instead of manually logging food items, users can take a photo of their meal, send it to the NutriScan Telegram bot, and instantly receive calorie and macro-nutrient information. The system is designed specifically for Indian cuisine, making it useful for home-cooked meals and regional dishes that are often missing from generic nutrition apps.

What NutriScan Does

• Detects food items from images using computer vision

• Recognizes 80+ Indian food dishes

• Provides estimated nutritional information such as:

  • Calories

  • Protein

  • Carbohydrates

  • Fats

  • Fiber and key minerals

• Works directly inside Telegram with no extra app or setup

How It Works (High Level)

1. The user uploads a food image to the Telegram bot

2. The backend processes the image using an AI model

3. Detected food items are mapped to nutrition data

4. A clean and readable nutrition summary is sent back to the user

All processing is automatic and requires no manual input.

Computer Vision (Food Detection)

NutriScan uses YOLOv8, a state-of-the-art object detection model, to identify food items from images.

• Trained and fine-tuned on 4,000+ images of Indian food

• Supports 80+ Indian food classes

• Can detect multiple food items in a single image

• Outputs bounding boxes and food labels (e.g., roti, dal, butter chicken)

The model was converted to ONNX format and runs efficiently in a Node.js backend using ONNX Runtime, allowing fast and production-ready inference.

Nutrition Intelligence

Once food items are detected, NutriScan retrieves nutrition data using the USDA FoodData Central API.

• Maps detected foods to standardized nutrition entries

• Extracts key nutritional values such as calories, macros, and minerals

• Filters and simplifies technical nutrient names into user-friendly labels

• Limits output to meaningful information to keep responses concise

Telegram Bot

• Built using the Telegram Bot API with a webhook-based architecture

• Users can start by sending a greeting or uploading an image

• Shows real-time “typing” indicators while processing

• Handles errors gracefully with clear, friendly messages

• Responses are formatted to respect Telegram message limits

Use Cases

NutriScan is useful for:

• Understanding what you’re eating

• Tracking nutrition without manual food logging

• Making healthier food choices

• Analyzing Indian meals that are not well supported by global apps

Limitations & Future Improvements

NutriScan is under active development and not 100% accurate yet.

• Some dishes may be misclassified due to visual similarity or image quality

• Portion size estimation is not currently supported

Planned improvements include:

• Portion size estimation

• Multi-food plate analysis

• Personalized dietary insights

• Larger and more diverse food dataset

• Expansion beyond Indian cuisine

If you liked this project, an upvote would mean a lot! Follow me for more projects like this.