Bulk OCR

Reading and consolidating text from multiple sources

In one case, I was assembling all the text and WhatsApp messages exchanged with a rogue trader. I have a Python program that whizzes through the screenshot of every interaction, reads all the text on each and assembles them into a complete dated history both in Spreadsheet form, and into a separate document that can be ingested into Google NotebookLM.

Once in NotebookLM, I was able to add in a complete email history of all discussions alongside the text messages, documents exchanged and over 25 audio transcripts. The resulting NotebookLM could then answer any question (in English) that related facts from 5 years of history, including summarising timelines, documenting excuses given for non-performance and records of all financial amounts cited along the way. All with the ability to link back to the source facts if there was any dispute.

This gave my client super powers in the eyes of her legal team fighting her case.

The separate we had for some journalist friends was to be at first pleased that the US Dept of Justics released the first block of evidence for the Epstein case in August 2025. However, it arrived in the form of nearly 23,000 one-page photographs of text from court proceedings and supporting newspaper, book and written documents. I have another program that transcribed all the text from every page and output them all into separate, searchable text documents.

At the time, some 444 of the evidence photo's submitted into the public domain evidence were of book or newspaper pages, which the AI refused to process on copyright grounds. So I built another program using an old Optical Character recognition package to process those pages.

The end result was that three sets of journalists had the ability to search for names and timelines from the very inconveniently formatted evidence. Personally, I have no interest in the content and have sat out of more recent 360GB dumps of what was badly redacted and I suspect dubious content.

If you have volumes of photographs or screenshots of text documents, I should be able to read them efficiently. 23,000 photos cost around £35 to process, and around 500 text message screen shots around 28p. The main time investment is in setup and ensuring you receive quality results.