Lumina

Lumina is a movie discovery engine that searches by meaning rather than keywords. Movie metadata is mapped into a 384-dimensional vector space with Sentence-BERT (all-MiniLM-L6-v2), so a query like a described mood or theme returns conceptually related films, and the system works without any historical user data. Retrieval runs in two passes: a broad semantic pass over cosine similarity, then a deterministic title-boost layer that pins exact keyword matches to the top so the model cannot bury the obvious answer. The Streamlit frontend handles inference and state, while TMDB API integration fetches posters and trailers asynchronously. Embeddings are generated on first run to keep the repository light.