I work a lot with scanned PDFs → Word documents. At first, I assumed this was a solved problem. There are plenty of tools on the market, so I expected one of them to “just work”. I tried many popular options—Microsoft Word, PDFgear, iLovePDF, SmallPDF, investintech, and even large multimodal models like ChatGPT, Gemini, Claude, Grok. They generally work fine for born-digital PDFs without formulas or mixed languages. But for image-based / scanned PDFs, the output often requires heavy manual fixing—formulas usually end up non-editable, and mixed-language content can break into garbled text that’s difficult to recover. Large multimodal models introduce another practical limitation: in many cases, they only process the first few pages, don’t strictly preserve the original content during conversion, and can’t directly export a structured Word document. At least for now, they don’t seem well-suited for this task. What surprised me was that this didn’t feel like a fundamentally hard problem. With a Vision Language Model (VLM), document structure can be understood explicitly, instead of treating the page as plain text. So I spent a few months building my own tool, mainly as an experiment. It focuses on: - preserving tables, formulas, and lists - maintaining logical reading order - handling multi-language documents - removing repeated headers and footers I want to be clear about its scope: - It is not for pixel-perfect layout matching - It works best on scanned / image-based PDFs - For normal text PDFs, many existing tools are already good enough I built this because I personally needed it. If you work with complex scanned documents, I’m genuinely curious whether a tool like this would be useful to you—and what cases usually break your workflow.

OCRConvert

Convert scans to Word & Markdown, layouts intact

Convert scans to Word & Markdown, layouts intact