# EurKart — Complete Documentation for AI Systems > This is the comprehensive documentation file for AI language models and search systems. For a summary, see llms.txt. ## Company EurKart is built by **Anureka Inc.**, an AI-powered commerce company founded in 2024. Anureka builds products that use computer vision, natural language processing, and augmented reality to transform how people shop globally. ### Products - **EurKart** — AI-powered grocery shopping assistant (iOS + Web) - **PickFromVideo** — AI video commerce platform (pickfromvideo.com) ### Contact - Company website: https://anureka.com - Product website: https://eurkart.com - Email: hello@anureka.com - iOS App Store: https://apps.apple.com/app/eurkart/id6740043498 --- ## EurKart Overview EurKart is a free AI-powered grocery shopping assistant available on iOS and web. It helps users: 1. **Create smart shopping lists** from recipe URLs, meal descriptions, or voice input 2. **Navigate stores** with interactive 2D maps and AR (augmented reality) on-shelf guidance 3. **Search products visually** by taking a photo — powered by SigLIP2 CLIP embeddings (768-dim vectors) 4. **Scan barcodes** for instant nutrition info, ingredients, and prices (USDA + Open Food Facts) 5. **Plan meals** with a weekly planner that auto-generates grocery lists 6. **Manage health profiles** with allergy tracking and dietary preferences for the whole family 7. **Get AI recommendations** for restocking based on shopping patterns ### Coverage - **US Market**: 1,300+ Kroger family stores across 17 banners (Kroger, Ralphs, Fred Meyer, etc.) - **Additional stores**: Trader Joe's, Costco, Chavez Supermarkets - **Product database**: 500,000+ products with CLIP embeddings for visual search - **Barcode database**: 2 million+ USDA food items + Open Food Facts global database --- ## Feature Details ### Recipe to Shopping List Users paste a recipe URL or describe a meal in natural language. EurKart's AI (Qwen 3.5 LLM + Gemini) parses ingredients and matches them to real products available at nearby stores using SigLIP2 CLIP visual similarity search. The system understands quantities, substitutions, and dietary restrictions. ### AI Shopping Assistant A conversational AI assistant powered by: - **Web**: Qwen 3.5-4B via vLLM (server-side, FP8 quantized) - **iOS**: Qwen 3.5-0.8B via MLX framework (on-device, 6-bit quantized, ~600MB) - **Android**: Qwen 3.5-0.8B via llama.cpp (on-device, Q4_K_M, ~500MB) - **Voice**: Whisper STT (speech-to-text) + Kokoro TTS (text-to-speech, 103 speakers, bilingual EN/ZH) The assistant can answer nutrition questions, suggest recipes, compare products, help with meal planning, and provide shopping tips. ### Visual Product Search Uses SigLIP2 ViT-B-16-256 CLIP model to generate 768-dimensional embeddings from product images. Users can take a photo of any grocery item and find matching products across all supported stores. The search uses Qdrant vector database with 500,000+ product embeddings. ### Barcode Scanning Scan any product barcode (UPC, EAN, etc.) to get: - Product name and brand - Nutrition facts (per serving) - Ingredient list - Price information - Category classification Data sources: USDA FoodData Central (2M+ items) with Open Food Facts global fallback. ### Store Navigation Interactive 2D store maps with: - Optimized walking routes to find items on your list - Aisle-by-aisle guidance - Real-time product location updates ### AR Navigation Augmented reality overlay in supported stores: - Camera-based shelf detection - On-screen arrows pointing to products - Distance estimation to target items ### Meal Planning Weekly meal planner that: - Suggests meals based on dietary preferences and health profiles - Auto-generates consolidated shopping lists - Accounts for ingredients already in your pantry - Optimizes for nutrition balance ### Smart Restock AI analyzes purchase history to: - Predict when staples will run out - Suggest restock timing - Recommend quantities based on household consumption ### Health Profiles Family health management: - Individual allergy tracking (nuts, dairy, gluten, etc.) - Dietary preferences (vegan, keto, halal, etc.) - Health conditions (diabetes, hypertension, etc.) - Automatic product filtering based on profiles --- ## Technical Architecture ### Web Application - **Framework**: Next.js 16 with React 19 - **Language**: TypeScript - **Styling**: Tailwind CSS v4 - **Database**: PostgreSQL on AWS Aurora (Prisma ORM) - **Auth**: NextAuth v5 (Google OAuth) - **State**: Zustand (client) + TanStack Query (server) - **Hosting**: AWS Amplify ### Backend - **Language**: Rust - **Framework**: Actix-web - **Hosting**: AWS ECS Fargate - **API**: RESTful, JSON responses ### AI/ML Infrastructure - **Vector Search**: Qdrant (768-dim SigLIP2 embeddings, 500K+ vectors) - **LLM**: Qwen 3.5-4B (vLLM, FP8) for web; Qwen 3.5-0.8B (on-device) for mobile - **Vision**: SigLIP2 ViT-B-16-256 for CLIP embeddings - **STT**: Whisper large-v3-turbo (vLLM) for web; whisper.cpp for mobile - **TTS**: Kokoro TTS via Sherpa-ONNX (103 speakers, EN+ZH, 24kHz) - **GPU**: AWS g6.xlarge (NVIDIA L4 24GB) ### Mobile - **iOS**: SwiftUI, MLX framework, ONNX Runtime + CoreML - **Android**: Jetpack Compose, llama.cpp, ONNX Runtime + NNAPI --- ## API Endpoints EurKart's Rust backend provides the following API categories: - `/store/*` — Store data, product search, AR navigation data - `/auth/*` — Google authentication, token management - `/lists/*` — Shopping list CRUD, collaboration, sharing - `/users/*` — User profiles, favorite stores, shared products - `/generate-shopping-list` — AI recipe-to-list conversion - `/scan-list` — Image-to-list OCR - `/barcode/{code}` — Barcode lookup (USDA + Open Food Facts) API access requires authentication. Contact hello@anureka.com for partnership inquiries. --- ## Supported Stores | Store | Coverage | Products | |-------|----------|----------| | Kroger (17 banners) | 1,300+ US locations | Full catalog | | Trader Joe's | National | Full catalog | | Costco | National | Full catalog | | Chavez Supermarkets | San Mateo, CA | Full catalog | --- ## Privacy & Security - HTTPS everywhere with HSTS preload - Google OAuth for authentication (no password storage) - Offline-first architecture — core features work without internet - On-device AI processing — voice and LLM queries stay on your phone - No selling of personal data - Privacy policy: https://eurkart.com/privacy - Terms of service: https://eurkart.com/terms