🍳 CookWise

A weekend vibe coding project for chatting with your recipes using local LLMs.

CookWise is a recipe OCR and RAG system that lets you upload recipe images/PDFs and have natural conversations with them using local AI models. Built with a focus on privacy and local processing.

🎯 What it does

• OCR Processing: Extract text from recipe images and PDFs
• Vector Storage: Store recipe embeddings for semantic search
• AI Chat: Ask questions about your recipes using local LLMs (Ollama)
• Web Interface: Clean Streamlit UI for easy interaction

🏗️ Architecture

High-Level Overview

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Recipe Files  │    │   OCR Pipeline  │    │  Vector Store   │
│  (Images/PDFs)  │───▶│  (Text Extract) │───▶│   (ChromaDB)    │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                                       │
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   User Query    │    │   RAG System    │    │   Local LLM     │
│  (Web Interface)│◀───│  (Retrieval)    │◀───│   (Ollama)      │
└─────────────────┘    └─────────────────┘    └─────────────────┘

Detailed Component Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Streamlit Web App                        │
├─────────────────────────────────────────────────────────────────┤
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │
│  │   Upload    │  │    Chat     │  │ Recipe Info │             │
│  │   Section   │  │   Section   │  │   Section   │             │
│  └─────────────┘  └─────────────┘  └─────────────┘             │
└─────────────────────────────────────────────────────────────────┘
                                │
                                ▼
┌─────────────────────────────────────────────────────────────────┐
│                        Core Processing                          │
├─────────────────────────────────────────────────────────────────┤
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │
│  │   OCR       │  │   Text      │  │   Vector    │             │
│  │ Processor   │──▶│  Chunker    │──▶│   Store     │             │
│  │             │  │             │  │             │             │
│  └─────────────┘  └─────────────┘  └─────────────┘             │
│         │                                 │                     │
│         ▼                                 ▼                     │
│  ┌─────────────┐              ┌─────────────┐                  │
│  │  Image      │              │   Recipe    │                  │
│  │ Processor   │              │  Retriever  │                  │
│  └─────────────┘              └─────────────┘                  │
└─────────────────────────────────────────────────────────────────┘
                                │
                                ▼
┌─────────────────────────────────────────────────────────────────┐
│                        AI Layer                                 │
├─────────────────────────────────────────────────────────────────┤
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │
│  │   LLM       │  │   Context   │  │   Response  │             │
│  │  Client     │──▶│  Formatter  │──▶│  Generator  │             │
│  │ (Ollama)    │  │             │  │             │             │
│  └─────────────┘  └─────────────┘  └─────────────┘             │
└─────────────────────────────────────────────────────────────────┘

Data Flow

1. Upload Recipe
   ┌─────────────┐
   │ Recipe File │
   │ (Image/PDF) │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ OCR Extract │
   │   Text      │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ Text Chunk  │
   │ (800 chars) │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ Vector DB   │
   │ (ChromaDB)  │
   └─────────────┘

2. Chat with Recipe
   ┌─────────────┐
   │ User Query  │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ Semantic    │
   │ Search      │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ Context +   │
   │ Query       │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ Local LLM   │
   │ (Mistral)   │
   └─────────────┘
           │
           ▼
   ┌─────────────┐
   │ AI Response │
   └─────────────┘

🚀 Quick Start

Prerequisites

• Python 3.11+
• Poetry (for dependency management)
• Ollama (for local LLMs)
• Tesseract (for OCR)

Installation

1. Clone and setup:

   git clone <your-repo>
   cd cookwise
   poetry install

2. Install Ollama:

   # macOS
   brew install ollama
   
   # Start Ollama
   ollama serve
   
   # Pull model
   ollama pull mistral:7b-instruct

3. Install Tesseract:

   # macOS
   brew install tesseract

4. Run the app:

   streamlit run app/main.py

🛠️ Tech Stack

Core Technologies

• Python 3.11 - Main language
• Streamlit - Web interface
• Poetry - Dependency management
• Loguru - Logging

OCR & Processing

• Tesseract - OCR engine
• PyPDF2 - PDF text extraction
• Pillow - Image processing
• pdf2image - PDF to image conversion

AI & ML

• Ollama - Local LLM server
• Mistral 7B Instruct - Language model
• Sentence Transformers - Text embeddings
• ChromaDB - Vector database

Development

• Ruff - Code formatting & linting
• Pytest - Testing
• Black - Code formatting

📁 Project Structure

cookwise/
├── app/                    # Streamlit web application
│   ├── main.py            # Main app entry point
│   └── ui/                # UI components
├── cookwise/              # Core library
│   ├── ocr/               # OCR processing
│   │   ├── image_processor.py
│   │   └── pdf_processor.py
│   ├── embeddings/        # Text processing & storage
│   │   ├── text_chunker.py
│   │   └── vector_store.py
│   ├── rag/               # RAG system
│   │   ├── retriever.py
│   │   └── llm_client.py
│   └── utils/             # Utilities
│       └── file_utils.py
├── data/                  # Data storage
├── tests/                 # Test files
├── pyproject.toml         # Poetry configuration
└── README.md              # This file

🎨 Features

OCR Capabilities

• Image Support: JPG, PNG, TIFF, BMP
• PDF Support: Direct text extraction + OCR fallback
• Multi-page: Handles multi-page documents
• Error Handling: Graceful fallbacks for different formats

AI Chat Features

• Context-Aware: Uses retrieved recipe context
• Local Processing: No data sent to external APIs
• Multiple Models: Support for any Ollama model
• Conversation History: Maintains chat context

Vector Storage

• Semantic Search: Find recipes by meaning, not just keywords
• Metadata Storage: File info, chunk indexing
• Scalable: ChromaDB handles large recipe collections

🔧 Configuration

Environment variables (optional, all have defaults):

# Ollama Configuration
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=mistral:7b-instruct

# Vector Store Configuration
VECTOR_STORE_PATH=./data/vector_store
EMBEDDING_MODEL=all-MiniLM-L6-v2

# OCR Configuration
TESSERACT_CMD=/usr/local/bin/tesseract

---

Built with ❤️ during a weekend vibe coding session