llm_aided_ocr

Why we included this project

Anyone who works with digitized archives or scanned correspondence knows OCR rarely comes out clean. This project splits the work: Tesseract does the initial extraction, then an LLM goes over the result to fix typos, reconnect words that got split apart, and restructure the page into readable markdown. You can run it against a local model when the material is sensitive, or point it at OpenAI or Anthropic when you would rather not manage hardware. Long documents stay coherent because the text is broken into overlapping chunks, and options to suppress headers and page numbers plus a built-in quality check give you control over the final file. For teams that regularly process legal records, old letters, or other scanned PDFs, it saves the kind of tedious cleanup that would otherwise fall to a person.