Click any title you've already loved on the map — scroll down for picks that feel the same. Search a name or franchise (try "james bond" or "agatha christie") to jump straight to it.
Embeddings: each title's name + year + format + mood/genre tags is run through Xenova/all-MiniLM-L6-v2, producing a 384-dimensional vector. Era and format-echo tags are stripped from the embedding text so clusters form around genre/feel rather than decade.
Layout: classical multidimensional scaling on the N×N cosine-distance matrix. The two coordinates you see are the top two eigenvectors of the double-centered squared-distance matrix.
Clusters: k-means with k=8 in the full 384-d space (not the 2D projection). Cluster names come from a curated template matched to the dominant tags of each cluster — so 'Sci-fi & space' isn't an arbitrary label, it's whatever fits the cluster's tag fingerprint best.
Neighbors vs visual distance: the list of neighbors is computed by cosine similarity in the original 384-d space. The 2D map is a projection — a lossy one — so two titles that are very close in 384-d space can still land far apart on screen, in different cluster colors. Clusters are a separate hard-assignment, so a title near a cluster boundary will have neighbors spread across colors. Trust the percentage in the list more than the visual distance on the map.
Why baked-in: N points × 384 dims is a lot for the browser to chew through. MDS, k-means, naming, and neighbor lists are pre-computed by scripts/build-layout.mjs and shipped as a static JSON file. The browser just renders.
Data sources: The catalog is hand-curated. Cast, director, official synopses, ratings, and runtime supplemented from TMDb. Streaming availability data, when shown, from Watchmode. This product uses the TMDB API but is not endorsed or certified by TMDB.