PantryPilot

About

NLP-based recipe recommendation system that matches grocery and pantry ingredients to recipes through a vectorized similarity search.

Designed for:

• Speed
• Explainability (why recipes are suggested)
• Food-waste reduction

Problem

• Choosing what to cook with limited ingredients is oftentimes difficult and can lead to major food waste
• Although recipe platforms provide great recommendations to users, it often makes the assumption that they will shop for missing ingredients, instead of based on what people already have

Solution

Model recipe selection through:

• Pantry items - treated as queries
• Recipes - treated as documents
• NLP vecotrization and similarity search
• Results optimized for pantry utilization

Features

• NLP-based recipe matching
  • Tokenizes and vectorizes recipe ingredients
  • Uses cosine similarity to rank recipes (based on overlap)
  • Scales to 250K+ recipes with vector search
• Explainability in recommendations (provided)
  • Overlapping ingredients
  • Similarity score
  • Pantry utilization percentage
• Pantry utilization (optimized)
  • Recipes are ranked by similarity and efficiency (how they use available ingredients)

Architecture

• User pantry
• Text preprocessing
• TF-IDF vectorization
• Cosine similarity search
• Ranked recommendations (top similarity)
• Explainability and utilization scores

How to Run

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python main.py