Thanks for sharing this. I've had great fun emulating the general pattern and hammering out genuinely useful tools. I got a lot of inspiration from your notes on malleable software and resonant computing: virtualarchitectures.github.io/Browser-Tools/

I possibly have lower expectations.

One thing I'm curious about is whether we could run these kinds of workflows with much more modestly sized local models, given the growing attention given to processes like skills which seem to focus on provisioning context, while diminishing reliance on parametric knowledge. I'd love to see that!

Docling is also fantastic for #DigitalHumanities. It plugs in to spaCy which is great for natural language processing (NLP): spacy.io/universe/pro...

Previously I built a custom data processing pipeline using object character recognition (OCR) with Tesseract to extract data from thousands of scanned PDFs...the most closed open data. Details here: datastories.maynoothuniversity.ie?p=674 Docling now saves so much time and code in data preparation.

There tends to be a knee jerk response to posts mentioning GenAI on Bluesky. That's disappointing because Docling is a double-edged sword...perfect for digital #counter-practice. I use it for liberating data that government agencies have inconveniently hidden in open sight within thousands of PDFs.