This project enhances Tesseract OCR output by using large language models to correct errors, restructure text into clean markdown, and assess output quality. It converts scanned PDFs to images, runs OCR, then applies LLMs (local or API-based including OpenAI and Anthropic) for correction and formatting. The system supports both local GPU-accelerated LLMs and cloud providers, with smart chunking for efficient processing.