Comment by sieve

I am continuing with my proofreading and language learning efforts and have been working on tooling for it.

= Proofreading =

https://github.com/adhyeta-org-in/adhyeta-tools

provides image extraction from PDF, OCR as well as a basic but nice proofreading web-ui.

Qwen 3/3.5 is good enough for OCR on books in Indic scripts. So that is what I am using. But you can configure the model that you want to use.

I may add a tesseract back end as well if necessary.

= Language Learning =

I have tried a few parallel text readers and was not satisfied by any of them. My website (https://www.adhyeta.org.in/) had a simple baked-in interface that I deleted soon after I developed it. However, this weekend, I sat down with Claude and designed one to my liking. I also ported the theming and other goodies from the website to this local reader. This will serve as a test bed for the Reader on the website itself.

LLMs now produce wonderful translations for most works. You can take an old Bengali book, have Claude/Gemini OCR a few pages and then also have it translate the content to English/Sanskrit. Then load it into the Reader and you are good to go!

The Reader, I will release this month. Claude is nice, but I do not like the way it writes code. It often misses edge cases and even some basic things and I have to remind it to do that. So I want to refactor/rearrange some stuff and test the functionality end-to end before I put it online.