Using AI to Digitize the Challis Index
I have been working with the Gemini AI to translate handwritten Portuguese documents and have had about 95% success rate. The two main issues are reading numbers and the choice of words used does not necessarily reflect 200 year old Portuguese. So I decided to see if it works for the Challis Index. I took its abbreviation file and converted it to a text document. I then uploaded the first 23 Challis pdf files which consisted of about 230 index card images.
It took about five minutes to convert it to digital file. I found only three minor errors. . . changing Abell to Abill.
Attached is the end results.
63 Views
I'I'in the process of setting up a major AI pipeline of historical document processing. The key is multi stage multi modal.
Sourcing, cataloguing, referencing, OCR, translation and final processing are completely seperate problems :-(.
There are tools such as have been mentioned, along with brute strength via LLM, but these can all get VERY expensive if applied bluntly at volume.
I'm setting up a multi level approach using cheaper options first, assessing, and triaging to selectively higher process based on results.