🍽️ NLP for Recipe Understanding & Voice Assistant

This project applies cutting-edge Natural Language Processing techniques to the RecipeNLG dataset, with the goal of building an intelligent, voice-interactive cooking assistant.

📌 What We Did

• Data Inspection & Preprocessing: Explored and cleaned the RecipeNLG dataset to extract meaningful fields (ingredients, instructions, titles).
• Recipe Classification: Trained models to classify recipes by type or cuisine.
• Embeddings + FAISS: Generated dense vector embeddings and built a fast retrieval system using FAISS for similarity search between ingredients, titles, and instructions.
• Named Entity Recognition (NER): Extracted structured entities like ingredients and cooking actions from raw recipe text.
• Voice Interaction:
  • Speech-to-Text (STT): Converts spoken queries into text.
  • Text-to-Speech (TTS): Synthesizes spoken responses from generated answers.
• Retrieval-Augmented Generation (RAG):
  • Used a local Mistral model enhanced with relevant context from FAISS to answer cooking-related questions such as:
    • "What can I cook with chicken and spinach?"
    • "Give me instructions for making pancakes."
  • Answers are provided back via TTS for a full voice experience.

📁 Notebooks

• NLP_Recipes_Final.ipynb
  Full pipeline covering all major NLP tasks: preprocessing, classification, NER, embeddings, FAISS, and integration with RAG, STT, and TTS.

• POC.ipynb
  A proof-of-concept notebook that demonstrates a complete voice-based assistant built using the components developed in the main notebook. It showcases real-world use cases like ingredient-based recipe suggestions and vocal instruction delivery.

🎯 Goal

To create a smart cooking assistant that understands spoken questions, retrieves relevant recipe knowledge, and speaks back useful, context-aware answers — merging traditional NLP with real-time, user-friendly interaction.