CoolFace
Apppublic

davanstrien/open-library-atlas

sourceHugging Faceupdated 2mo agoView on Hugging Face
5likes
App README

🗺️ Open Library Atlas

Interactive embedding visualization of 2 million book titles from Open Library — "one web page for every book ever published."

Each point is a book title, colored by subject category (Fiction, History, Science, Religion, and more).

How it was built

This atlas was built entirely on Hugging Face infrastructure using the new Storage Buckets feature:

  1. 1.Prep (24s): DuckDB queried 40M Open Library works remotely via hf://, filtered to 2M books with subjects, and wrote enriched parquet to a Storage Bucket
  2. 2.Build (~40 min): An HF Job on A100 computed text embeddings + GPU-accelerated UMAP (via cuml.accel), writing results to the same bucket
  3. 3.Serve: This Docker Space mounts the bucket read-only and runs Embedding Atlas in server mode with DuckDB

No uploads, no CORS workarounds — the bucket is the shared storage between Jobs and Spaces.

Features

  • —2,000,000 embedded book titles with UMAP projections
  • —Color by category: Fiction, History, Science, Religion, Biography, and more
  • —Interactive scatter plot with WebGPU acceleration
  • —Real-time search and filtering
  • —SQL queries via DuckDB server mode

Categories

Books are categorized by subject into: Fiction, History, Science, Religion, Biography, Literature, Tech & Engineering, Art & Design, Law & Politics, Business & Economics, Education, Philosophy, Medicine, Children, Travel, Society, Military, Music, Food & Agriculture, Sports.

Credits