Project 01 / In development
Mediur
Can a book and a film leave you with a similar feeling? Mediur explores recommendations built around emotional tone, pacing, themes, and atmosphere.
Implemented so far
- FastAPI routes for media items and similarity retrieval
- Rule-based vibe profiles
- SentenceTransformer embeddings stored with pgvector
- Cosine-distance ranking
- TMDB movie catalogue search
Next questions
- How should selected catalogue results be imported?
- How can profile extraction capture more nuance?
- Which examples would demonstrate useful cross-media matches?
- How should recommendation quality be evaluated?
Status recorded 9 September 2026 from the repository. This is an early implementation; recommendation quality has not yet been established here. Movie search does not currently import records.
Development journal
All entries →Foundations / Recommendations
Why I’m building Mediur
The idea, the first implementation, and the questions that still need evidence.