Latent Space Explorer

PROJECT_ABSTRACT

Latent Space Explorer embeds arXiv abstracts offline with a MiniLM sentence transformer, projects the 384-dimensional vectors down to 3D through truncated SVD and UMAP, and renders the result as a navigable point cloud in a single WebGL draw call. All expensive work happens once in a reproducible three-stage pipeline, so what ships is a purely static site: under a megabyte of binary buffers. Free-text queries are embedded in the browser with transformers.js and ranked against int8-quantized vectors inside a Web Worker.

KEY_CAPABILITIES

• ▶Single draw call for the whole cloud. a custom GLSL shader handles colour filtering and selection, so frame cost tracks rasterized pixels rather than point count.
• ▶In-browser semantic search: transformers.js embeds the query with the same MiniLM model used offline and a Web Worker brute-force scans the int8 codes in milliseconds.
• ▶One truncated SVD basis serves double duty — it feeds UMAP and produces the 128-D search codes, so runtime queries land in exactly the same space as the offline corpus.
• ▶Nearest-neighbour lookup per-archive filtering from the legend and a layout that shifts from hover-to-inspect on desktop to tap-and-sheet on mobile.