Snap Cook: An AI-Driven Food Recognition and Personalized Recipe Recommendation System

The rapid growth of digital food platforms has reshaped how individuals discover and prepare meals, yet most existing recipe systems remain confined to text-based keyword search and lack meaningful visual understanding or adaptive personalization. This paper presents Snap Cook, an AI-driven cooking assistance platform that identifies food items directly from images and delivers personalized recipe recommendations through an integrated pipeline combining computer vision, hybrid machine learning recommendation logic, and rule-based dietary filtering.
The system employs a fine-tuned Convolutional Neural Network (CNN) trained on over 12,000 labeled images from the Food-101 dataset augmented with custom samples, achieving an image recognition accuracy of 94.2%. A content-driven recommendation engine built with Scikit-learn matches recognized dishes to a curated recipe catalog, while a rule-based dietary filtering layer handles vegetarian, vegan, gluten-free, and allergen-based constraints with 97.1% filtering accuracy. The full-stack implementation uses Flask (backend), React.js (frontend), PostgreSQL (database), and OpenCV (preprocessing), containerized via Docker for reproducible deployment.
Experimental evaluation across five functional modules yields a system-level mean accuracy of 92.1% and an F1-score of 92.1%, outperforming baseline text-search platforms on visual discovery and personalization benchmarks. Limitations including dataset scope, real-time nutritional API integration, and offline accessibility are acknowledged, and a roadmap for future enhancement through generative AI, augmented reality guidance, and multilingual support is presented.