Hey everyone ![]()
I built Docling Studio, an open-source visual inspection layer for Docling.
The problem: if you’ve used Docling, you know the extraction engine is powerful — but validating outputs means digging through JSON and mentally mapping bounding box coordinates back to the original pages. No visual feedback loop.
What Docling Studio does:
-
Upload a PDF, configure your pipeline (OCR engine, table extraction, enrichment)
-
Run the conversion
-
Visually inspect every detected element — bounding boxes overlaid on original pages, element types, content preview on click
-
Two modes: local (embedded Docling) or remote (Docling Serve)
Stack: Vue 3 / TypeScript + FastAPI / Python, fully Dockerized (multi-arch), 180+ tests.
Why it matters for RAG workflows: without seeing what Docling extracts, it’s hard to evaluate chunk quality before feeding your pipeline. Docling Studio gives you that feedback loop.
Repo: github.com/scub-france/Docling-Studio
Try it: Docling Studio on HF Spaces
The project is part of the Docling ecosystem (Linux Foundation AI & Data). Feedback, issues, and contributions welcome — especially interested in hearing what features would help your document AI workflows.