pankajsanger/amazon-rag-assistant

This project is a tool that scrapes product listings and customer reviews from Amazon, stores them in a local database, and then lets you ask plain English questions about what it collected. It answers those questions using a technique called retrieval augmented generation, where the system searches through the stored data for the most relevant pieces before asking a language model to write an answer grounded in that information, all wrapped in a simple web interface built with Streamlit. You can feed it a single product page link, upload a spreadsheet with a list of links, or run a live keyword search on Amazon. It collects two kinds of information side by side, product details like title, price, and rating, and customer reviews with their author, star rating, date, and text. Everything is saved into a local SQLite database, so scraping the same product again updates the existing entry instead of creating a duplicate. Products and reviews are each turned into their own searchable index using a vector database called Chroma, and a self query retriever lets the language model translate a question like reviews rated below three stars into the right filters automatically, rather than requiring you to write any filtering logic yourself. A rebuild index button in the app re-embeds new data at any time without needing to restart the whole program. To log into Amazon without dealing with captchas or two factor prompts, the project reuses a saved browser session stored as cookies rather than automating a real login each time. Under the hood it relies on Selenium and BeautifulSoup for browser automation, OpenAI for both the embeddings and the answering language model, and LangChain to tie the retrieval and generation steps together. The author describes this as a personal, educational project for learning how a full retrieval augmented generation system fits together end to end, and notes that anyone using it should respect Amazon's terms of service and robots.txt rules before scraping at any scale. It requires Python, Google Chrome, and an OpenAI API key to run.