Show HN: I'm running ML models for PDF layout analysis entirely in the browser

Client-side document structure detection running entirely in your browser. Uses ONNX Runtime (WebAssembly/WebGPU) to run deep learning models locally -- no server, no API keys, your documents never leave the tab.

Layout detection -- identifies text blocks, titles, tables, figures, headers, footers, and other structural elements on each page.

Table structure recognition -- detects rows, columns, column headers, and spanning cells within tables identified by the layout model.

This runs deep learning inference directly in your browser tab. The models are ~60 MB each and inference is CPU/GPU intensive. Expect high memory usage and your tab may become unresponsive during analysis, especially on mobile devices or older hardware.

The layout analysis plugin works with React, Svelte, and Vue. Everything above is built with @embedpdf/plugin-layout-analysis and runs on the headless @embedpdf/core plugin system.