Javascript alternative?

[vtempest]

Check out airesearch.js.org for a lot of similar pdf and html to markdown tools. It'd be great for a serverless and frontend js version so I can integrate the algorithms

[cr4yfish]

If anyone wants to team up, I'd be up to creating a Typescript port of this for React.

[vtempest]

Sounds good. It can also have a cite like airesearch.js.org

[mtin79]

• 1 for javascript / typescript

[cr4yfish]

I created a repo for it: docling-js

[drmingler]

This will be really interesting, you can leverage this backend API which I built for Docling here docling-api

[vtempest]

We want a tool that can run in serverless Node.js not something that needs to have a python server running.

It could be based on pdfjs like this https://github.com/vtempest/ai-research-agent/blob/master/src/extractor/url-to-content/pdf-to-content.js#L21

[Anatolii-Ovcharuk]

To implement a serverless PDF-to-HTML/Markdown converter in Node.js, the structure from your provided code is a great starting point. Using pdfjs-dist (the npm package version of pdf.js) would enable a streamlined integration, avoiding Python dependencies.

Here are key components to refine and implement for serverless use in Node.js:

PDF Loading and Parsing: Use pdfjs-dist for loading PDFs in Node.js. With resolvePDFJS, we can dynamically import or configure pdf.js in a way that suits serverless environments.

Options Handling: Set up options such as addHeadingsTags, addPageNumbers, and moveFootnotes to structure the output as HTML, aligning the tool to user needs without extra dependencies.

Processing PDF Text Content: Parse the text content with getTextContent, then analyze line heights, as in your original example, to infer headings and footnotes. Use helper functions to organize text into chunks based on styles (headings, paragraphs, etc.).

Document Structure & Output: Construct the HTML by mapping parsed content to HTML tags and structuring it with options for line breaks, headings, and footnotes.

Adding PDF Metadata: Leverage pdf.js’s getMetadata method to extract metadata for authorship and citation.

[PeterStaar-IBM]

@vtempest @drmingler @cr4yfish This is not an issue we want to address here, but we do love the idea of it. Hence, we think it is great to convert it as a discussion.

[Anatolii-Ovcharuk]

Airesearch.js.org could indeed be a good fit for creating a serverless, frontend JS tool for handling PDF and HTML to Markdown conversions. This would let you integrate lightweight algorithms directly into your frontend without relying on server-side processing, making it faster and potentially more secure. Do you have any specific frameworks or libraries in mind for this integration, like React or Next.js?

[vtempest]

What's missing from airesearch.js.org extraction is the special handling like tables eq images from pdf. Can you please translate this