terror-reco 2024
projects
A horror movie recommendation engine powered by semantic search, sentence-transformers, and the OMDb API. Describe a mood, scene, or vibe in plain text and get matched with horror movies that fit — no categories, no keyword lists.
For example:
- “a slow-burn psychological thriller set in a remote cabin”
- “found footage alien abduction at night”
- “gothic vampire romance in Eastern Europe”
How it works
A FastAPI app routes each query to one of four recommendation strategies, then applies filters and stochastic sampling so results stay fresh on every search:
-
Semantic Search — sentence-transformer embeddings (
all-mpnet-base-v2) match your text against movie plots by meaning, not keywords. - Unified — blends semantic scores with MMR diversity for varied results.
- TF-IDF Similarity — scikit-learn cosine similarity over OMDb plots.
- Keyword Match — OMDb title search ranked by IMDb rating.
Features
- Free-text mood/scene search over a pre-built horror corpus
- Advanced filters: year range, minimum IMDb rating, type, English-only
- Movie detail modal with director, cast, runtime, awards, and a direct IMDb link
- Like / dislike feedback stored per user for future personalisation
- User accounts with Argon2 password hashing and CSRF protection, plus search history
- Stripe “buy me a coffee” checkout with webhooks
- A dark, responsive horror-themed UI
- Dockerised with a CI/CD pipeline (lint, type-check, test, build)