Amazon Reviews RAG (100K)

This project is a question-answering demo over 100K Amazon product reviews, using a simple retrieval‑augmented generation setup (vector search, BM25, and reranking) with a Gradio UI.

---

What it does

• Builds a FAISS index over 100K Amazon reviews
• Runs hybrid retrieval (vectors + BM25) with a cross‑encoder reranker
• Uses an OpenAI chat model to answer questions based only on retrieved reviews
• Exposes a small Gradio app to ask questions about the reviews

---

Setup

Install dependencies (Python 3.9+ recommended):

pip install langchain-core langchain-community langchain-openai \
    sentence-transformers faiss-cpu rank-bm25 gradio datasets tqdm python-dotenv

Prepare data and build the index:

python -m src.ingest
python -m src.build_index

---

Run

Start the Gradio app:

python -m src.app_gradio

Then open the URL printed in the terminal and ask questions like “What do customers say about battery life?”.