This morning's fun has been working on a Golang CLI (it has better PDF processing options then Deno) to interface with https://ollama.com/benhaotang/Nanonets-OCR-s via local Ollama. It's a bonkers good OCR model based on a specially trained Qwen.

This is a page from equimundo's "State of American Men", and this is the result: https://ray.so/kPoztjf

(My Chrome alt txt xtsn also did πŸ‘πŸΌ extracting meaning from the πŸ“ˆ)

Page 6 results were bonkers cool, too.

Heat wave kicks in tomorrow, so shld have code up by then.

oh, if the model is in memory it takes just a couple seconds, too.