cookoo – Personalized RAG-Powered Recipe Advisor Chatbot

A modern, full-stack personalized recipe advisor chatbot that helps users discover, adapt, or create recipes based on ingredients they have, dietary preferences, time constraints, servings, and cooking style.

Powered by Retrieval-Augmented Generation (RAG) with local embeddings and LLM inference — built as a portfolio project to showcase applied AI engineering, RAG pipelines, and clean full-stack development.

https://github.com/elhamfo/cookoo

Features

• Semantic Recipe Search
  Uses sentence-transformers embeddings + FAISS for intelligent, context-aware retrieval over thousands of recipes

• Retrieval-Augmented Generation (RAG)
  Combines retrieved recipes with user preferences to generate personalized, practical recipe suggestions

• Conversational Chat Interface
  Message bubbles, loading states, auto-scroll, and natural back-and-forth interaction

• Dietary & Constraint Filters
  Vegan, gluten-free, low-carb, quick meals, servings count, ingredient exclusions, etc.

• Fully Local Inference Option
  No API keys required when using Ollama (Llama 3.2) — privacy-focused and offline-capable (dev only)

• Clean, Responsive UI
  Modern design with Tailwind CSS v4, lucide-react icons, and smooth interactions

• API Documentation
  Interactive Swagger/OpenAPI docs for the backend endpoints

Tech Stack

Backend

• Python 3.11+
• FastAPI (API framework)
• LangChain + langchain-ollama / langchain-openai
• sentence-transformers (embeddings)
• FAISS (vector store)
• Ollama (local LLM – Llama 3.2 3B / 1B)

Frontend

• React 18+ (Vite)
• Tailwind CSS v4
• lucide-react (icons)
• Axios (API communication)

Data

• Public recipe dataset (~13k recipes): https://github.com/josephrmartinez/recipe-dataset

LLM Options

• Local: Ollama (Llama 3.2) – dev only
• Cloud fallback: OpenRouter free tier (auto-routes to available open models)

Quick Start (Local Development)

Prerequisites

• Python 3.11+
• Node.js 18+
• Ollama installed & running (for local LLM: download from ollama.com, run ollama pull llama3.2)

Clone repo

git clone https://github.com/elhamfo/cookoo.git cd cookoo

Backend setup

python -m venv venv source venv/bin/activate # Windows: .\venv\Scripts\activate pip install -r requirements.txt

Prepare recipe data & vector index (run once)

python prepare_data.py

Start backend

python app.py

→ http://localhost:8000/docs (Swagger UI)

In another terminal → Frontend

cd frontend npm install npm run dev

→ http://localhost:5173

Test the app locally:

• Open http://localhost:5173 in your browser
• Ask for recipes based on ingredients, preferences, or constraints
• Try dietary filters and local Ollama mode (if Ollama is running)

Why This Project?

As an AI engineer, I created this project to demonstrate:

• RAG Pipeline
  End-to-end retrieval + generation using local embeddings and FAISS

• Local LLM Integration
  Ollama for privacy-focused, offline-capable inference

• Conversational AI UX
  Clean chat interface with filters, state management, and responsive design

• Full-Stack Production Skills
  FastAPI API design, React/Vite frontend

• Security & Best Practices
  Input validation, no hardcoded secrets, proper CORS, free-tier cloud deployment

Improvements in Progress

• Multi-turn memory & conversation history
• Recipe image generation/visualization
• User profiles & saved favorites
• Advanced personalization & ranking

Contact

Elham Fo
📧 elham.fo@gmail.com

Open to collaborations, feedback, and discussions on applied AI engineering, RAG systems, conversational AI, and full-stack ML applications!