Context-Aware Recipe Generator

Adapt-a-dish is an AI-powered recipe transformation assistant that customizes recipes based on dietary preferences, skill level, and serving size. Whether you're vegan, gluten-free, or exploring healthier options, Adapt-a-dish adapts your favorite recipes into personalized culinary masterpieces.

It uses a fine-tuned Vision Transformer (ViT) model with LoRA (Low-Rank Adaptation) for image-based recognition and incorporates user-defined contexts (e.g., "vegan", "low-calorie") to customize recipe results. The app fetches recipe data using the Tavily search API and generates detailed recipes through a LLaMA model accessed via the Groq API. For customizing recipes, the system uses ChromaDB and Ollama for ingredient substitution and recipe transformation.

---

🚀 Features

• Image-Based Dish Recognition: Upload an image to identify a dish using a fine-tuned ViT model with LoRA.
• Text-Based Dish Input: Input a dish name directly for recipe generation.
• Context-Aware Filtering: Customize recipes with specific dietary contexts:
  • Vegan
  • Low-calorie
  • High-protein, etc.
• Recipe Generation: Generates recipes with ingredients and step-by-step instructions using Tavily's search API and LLaMA 3.1 provided by the Groq API.
• User-Friendly Interface: Simple and interactive frontend developed in Angular.

For Customizing Recipes:

• Ingredient Substitution: Uses ChromaDB for finding suitable ingredient replacements.
• Recipe Transformation: Ollama is used for transforming recipes based on dietary preferences and skill level.

---

📚 How It Works

1. User Input:

   • Provide the dish name, dietary preferences (e.g., vegan, gluten-free), serving size, and a recipe to adapt.

2. AI Query:

   • The system uses the provided dish name and preferences to retrieve substitution ideas and context.

3. Recipe Transformation:

   • The recipe is adjusted with dietary-friendly ingredients and cooking instructions tailored to the skill level.

4. Output:

   • The system generates a personalized recipe with detailed ingredients and instructions.

---

🛠️ Tech Stack

• Backend:
  • FastAPI, Ollama, ChromaDB, PyTorch, LangChain, Transformers, Torchvision
• Frontend:
  • Framework: Angular
• APIs:
  • Tavily API: For finding relevant recipes.
  • Groq API (LLaMA): For generating detailed recipe content.
• Dependencies:
  • torch, transformers, langchain, fastapi, pydantic, torchvision, ollama, chromadb, etc.

---

📋 Installation

1. Clone the Repository:

   git clone https://github.com/Atul-AI08/Context-Aware-Recipe-Generator.git
   cd Context-Aware-Recipe-Generator

2. Finetune ViT: Ensure Python 3.8+ is installed along with required libraries, then run:

   cd backend/vit/
   python finetune_vit.py

   Dataset Used: The-massive-Indian-Food-Dataset

3. Set Up APIs: Create API keys for Tavily and Groq

4. Run the Backend:

   cd backend/vit/
   python dish_detect.py
   cd ../generation/
   ollama serve
   python main.py

5. Run the Frontend: Navigate to the Angular project directory and start the development server:

   cd frontend
   npm i
   ng serve

---

🌐 Example Images

Input and Recognition

Example of image upload for dish recognition.

Context Selection

Interface for selecting dietary preferences.

Recipe Output

Generated recipe with ingredients and instructions.

---

📄 License

This project is licensed under the MIT License. See the LICENSE file for more details.

---

🤝 Contributing

Contributions are welcome! Fork the repository and submit a pull request for review.