🍳 Hurry up, I'm hungry! - Culinary Recipe Generator

This project uses AI to generate novel and plausible culinary recipes based on a user-provided list of ingredients.

⚙️ How It Works:

A two-stage process:

1. Technique Prediction (Word2Vec):

   • Analyzes input ingredients.
   • Predicts suitable cooking techniques (e.g., bake, boil) based on learned semantic relationships from the dataset.

2. Recipe Generation (Fine-tuned GPT-2):

   • Takes the original ingredients and the predicted techniques as input.
   • A fine-tuned GPT-2 model, trained on a vast corpus of recipes, generates the complete recipe steps using special tokens ([INGREDIENTS], [TECHNIQUES], [STEPS]) for structured output.

📊 Evaluation

Generated recipes are evaluated against real recipes by:

1. Finding the most similar real recipes based on ingredient overlap (TF-IDF & Cosine Similarity).
2. Comparing generated steps to benchmark steps using:
   • ROUGE-L: Measures lexical overlap.
   • BERT Similarity: Measures semantic similarity.

(Dataset: Based on Food.com recipes from Kaggle)