LLM+RAG-Powered Product Recommendation System

LLM+RAG-Powered Product Recommendation System

BERT

embedding

LLM

Docker

Postgres

MLFlow

FastAPI

classifier

Key-words

Resources

GitHub

Overview

We download Amazon product reviews data from https://amazon-reviews-2023.github.io/, load them to a postgres database and develop a recommender system.

  • embed reviews using BERT, OpenAI embeddings or TF-IDF method

  • train a sentiment classifier on reviews

  • create a dashboard for data vis. and models monitoring

  • implement the recommendation system

  • implement RAG components

  • develop chatbot interface calling an agent transferring questions to LangChain chains for database-querying or to other utils

  • wrap up workflow with Docker

Approach

  • postgres for data storage

  • MLFLow for models registering

  • Docker for wrapping up workflow

  • FastAPI+Dash for dashboard

  • FastAPI+Dash for chatbot interface

  • spaCy+NLTK+TextBlob for NLP

  • BERT, XGBoost, sklearn for building models (sentiment classifiers)

  • PyTorch for tuning BERT components

  • LangChain for RAG components