Show HN: I'm running ML models for PDF layout analysis entirely in the browser (www.embedpdf.com)

A new project showcased on Show HN brings machine learning capabilities for PDF layout analysis directly to the browser, eliminating the need for server-side processing or API keys. Utilizing ONNX Runtime via WebAssembly and WebGPU, this tool performs client-side document structure detection that can identify critical components such as text blocks, titles, tables, and other structural elements without sending any data off the local device. This empowers users with greater privacy and control over their documents. The significance of this innovation lies in its ability to make advanced AI tools more accessible, particularly for applications in document analysis and data extraction. With the integration of PaddleOCR and PaddlePaddle, the system can also accurately recognize complex table structures, including rows, columns, and spanning cells. While the models are relatively lightweight at around 60 MB, users should be aware of the potential for high memory consumption during operation, which may lead to unresponsiveness, especially on mobile or older systems. This development marks a notable step in bringing sophisticated AI capabilities into mainstream web applications.