NextFlix is a fullstack AI-powered recommendation system for movies, TV shows, and anime. It uses semantic search via Pinecone and a Retrieval-Augmented Generation (RAG) pipeline with Llama 3 (via Groq) to deliver high-quality, natural language recommendations based on user prompts.
💡 What It Does
Semantic Search: Searches across 10,000+ movies, TV shows, and anime using vector embeddings in Pinecone.
Query Understanding: Uses Llama 3 to rewrite user prompts into structured semantic queries.
LLM-Generated Output: Ranks and formats results into clear recommendations via a second Llama 3 call.
Filter Support: Allows filtering by type (movie, TV, anime), genre, and more.
Natural Language Interface: Responds to open-ended prompts like "Give me 10 anime series about another world" with accurate, friendly results.
🛠️ How It Was Built
Backend:
Built with FastAPI and deployed on Render. A single endpoint handles the full RAG pipeline:
- Scrapes over 10,000 media entries from TMDB, including genre, rating, and metadata.
- Transforms user input into a semantic search query using Llama 3 (via Groq).
- Embeds the query using
llama-text-embed-v2and searches it in Pinecone.
- Passes results to Llama 3 again for clean natural language output.
- Ensures outputs are grounded in indexed content — no hallucinations.
Frontend:
Built with Next.js and styled using
assistant-ui's LLM template:- Chat-style interface with thread history and prompt input.
- Hosted on Vercel with a custom domain: nextflix.rohanrashingkar.com