CineSearch — AI Content Discovery Engine

Semantic search and Q&A across a movie/series catalog using Retrieval-Augmented Generation (RAG).

What It Does

• Ingest movie metadata and plot summaries as documents
• Chunk text with LangChain's RecursiveCharacterTextSplitter
• Embed chunks via sentence-transformers (all-MiniLM-L6-v2)
• FAISS vector index for sub-50ms similarity search
• Generate answers with Ollama (Llama3) grounded in retrieved context
• FastAPI streaming API with Server-Sent Events
• Docker Compose: one command local setup

Architecture

[Ingester] -> [Chunker] -> [Embedder] -> [FAISS Index]
                                          |
[User Query] -> [Retriever] -> [Grounded Generator] -> [Streaming Response]

Tech Stack

Python | LangChain | FAISS | sentence-transformers | Ollama (Llama3) | FastAPI | Docker

Quick Start

git clone https://github.com/neuralasmi/CineSearch
cd CineSearch
docker-compose up --build
# Visit http://localhost:8000/docs for interactive API

API Endpoints

• POST /ingest — Ingest documents into the vector store
• POST /query — Ask a question, get a streaming RAG response
• GET /health — Health check