HEB Product Retrieval and Reranking System

This project implements a hybrid retrieval and reranking pipeline for matching queries to product descriptions.
It combines OpenAI embeddings, FAISS (or NumPy fallback) for similarity search, and a cross-encoder model for reranking.
The system can be used for query–product retrieval tasks such as e-commerce search, information retrieval, and synthetic query benchmarking.

---

Features

• Text preprocessing and normalization (stopword removal, lowercase, alphanumeric filtering)
• Embedding of product metadata fields (title, brand, category, description) using OpenAI embeddings
• Construction of an inner-product (cosine similarity) index using FAISS or a NumPy fallback
• Query embedding and top-K retrieval
• Cross-encoder reranking using a transformer model (e.g., cross-encoder/ms-marco-MiniLM-L-6-v2)
• Batch processing of all queries to generate ranked product predictions
• JSON output compatible with evaluation pipelines (e.g., submission.json format)

---

Installation

Requirements

• Python 3.9+
• Dependencies:

  pip install -r requirements.txt
  
  
  

HOW TO RUN: /usr/local/bin/python3 PATH/semanticSearch.py --products [PATH_TO_PRODUCTS_FILE] --queries [PATH_TO_QUERIES_FILE]