CineSeek is a semantic movie search system that understands natural-language queries. It combines embedding-based retrieval, FAISS-powered candidate search, and an optional LLM agent for reranking and explanation.
The system keeps retrieval explicit and efficient: sentence embeddings, FAISS-based nearest neighbor search, and a strong frozen embedding baseline selected through evaluation. An optional LLM agent improves ranking and adds explanations, without replacing the underlying retrieval pipeline.
CineSeek is built for real, fuzzy movie queries — not just exact title matches. It connects natural-language intent to relevant results through semantic retrieval and optional agent-based refinement.
Queries and movies are mapped into a shared embedding space, so the system can retrieve relevant titles beyond exact keyword overlap.
FAISS provides low-latency nearest neighbor search, making the retrieval layer fast enough for interactive search experiences.
An optional LLM agent can refine vague queries, rerank retrieved candidates, and generate short explanations for the final results.
The query is embedded and matched against a FAISS index built from movie representations.
An LLM agent reorders the top candidates and helps resolve vague, broad, or ambiguous descriptions.
The system returns short explanations so the final recommendations feel more transparent and easier to inspect.