Netflix RecSys - By Rashmin Gajera

Project Prototype

---

🚀 Overview

Netflix RecSys is a modular, local-first Netflix-style movie recommendation system. It features:

• FastAPI backend
• Sentence-Transformers for embeddings
• ChromaDB for vector storage
• Reranking (cosine similarity, TensorFlow cross-encoder)
• LangChain for LLM explanations
• Simple HTML/JS frontend

---

🛠️ Tech Stack

---

Directory & File Structure Overview

Top-Level

• README.md
  This file. Project overview, file walkthrough, and credits.

• RUN_INSTRUCTIONS.md
  Step-by-step guide to running, training, and extending the system locally.

• docker-compose.yml
  Docker Compose setup for ChromaDB, API, and frontend services.

• data/

  • processed/sample_movies.jsonl
    Sample movie dataset (JSONL format) for ingestion and testing.

---

services/

api/

• requirements.txt
  All Python dependencies for the API service.

• app/

  • main.py
    FastAPI app entrypoint. Loads API routes and health check.
  • api.py
    Registers and includes all API routers.
  • routes/recommend.py
    /api/recommend/query endpoint: Accepts a query, retrieves candidates, reranks, and returns recommendations.

embeddings/

• embedder.py
  Embedding pipeline using sentence-transformers. Also provides a CLI for ingesting data into ChromaDB.
• requirements.txt
  Dependencies for embedding and ingestion.

retriever/

• chroma_client.py
  ChromaDB client for upserting and searching movie vectors.
• requirements.txt
  ChromaDB dependency.

reranker/

• model.py
  Lightweight in-memory reranker using cosine similarity.
• tf_crossencoder.py
  TensorFlow-based cross-encoder reranker (trainable, optional).
• requirements.txt
  Numpy dependency for reranking.

scraper/

• selenium_scraper.py
  Selenium-based IMDB plot fetcher (stub/example).
• requirements.txt
  Selenium and requests dependencies.

langchain/

• rag.py
  LangChain-based RAG pipeline for LLM-powered recommendation explanations.

frontend/

• index.html
  Simple HTML/JS frontend for demoing the recommendation API.

---

How It Works

1. Ingest Data:
   Use services/embeddings/embedder.py to encode movie plots and store them in ChromaDB.

2. API Service:
   services/api/app/main.py launches a FastAPI server exposing /api/recommend/query.

3. Retrieval & Reranking:

   • services/retriever/chroma_client.py retrieves candidates from ChromaDB.
   • services/reranker/model.py reranks candidates by similarity to the query.

4. Frontend:
   services/frontend/index.html provides a simple UI to test recommendations.

5. Optional Extensions:

   • services/reranker/tf_crossencoder.py: Trainable reranker.
   • services/langchain/rag.py: LLM-based explanations for recommendations.
   • services/scraper/selenium_scraper.py: IMDB plot scraping utility.

---

Contributing

This project is open for suggestions and improvements!
Feel free to open issues or pull requests.

---

Credits

Made by Rashmin Gajera.

---