DeepSeek-OCR-Web

Why we included this project

Most OCR tools stop at plain text; this one keeps going. It wraps the DeepSeek-OCR model in a React frontend and FastAPI backend, so you upload a PDF or image through the browser and get parsed output back without scripting the model yourself. Beyond text recognition it handles layout analysis, extracts tables and chart data, and converts documents to Markdown, which is what you want when reports and technical drawings need to flow into downstream pipelines. It also reads professional-domain drawings like CAD files and flowcharts, a niche generic OCR tools usually skip. The tradeoff is that it's self-hosted: you need a Linux machine with a GPU (at least 7 GB of VRAM), so it suits teams that already run their own inference infrastructure more than teams expecting a hosted API.