This content originally appeared on DEV Community and was authored by Adam LB
I’m excited to share a new project I’ve been working on: Documents-Parsing-Lab
This project is not a benchmark, but an exploration lab — built around a collection of Jupyter notebooks that test different AI models on various document structures. The idea is simple: let users see the parsing results directly, compare approaches, and understand where each model shines (and where it struggles).
What’s Inside?
Each notebook applies one or more models to different real-world document types, including:
- PDFs
- Tables
- Charts & Figures
- Complex layouts
Models Explored So Far
-
OCR & Vision Models
- Dolphin (ByteDance)
- Typhoon_OCR_7B
- MonkeyOCR
- PaddleOCR
-
Table Structure Recognition
- UniTable
- TableFormer
- Microsoft Table Transformer
(…and more to come!)
Why This Project?
There are lots of benchmarks out there — but in practice, people want to see how models handle real documents. This lab is meant to be a hands-on exploration space where you can:
- Run the notebooks yourself
- Compare outputs across models
- Understand limitations before using them in production
What’s Next?
More models and examples will be added over time. The goal is to keep expanding this into a growing resource for anyone interested in:
- OCR
- Document AI
- Structured data extraction
This content originally appeared on DEV Community and was authored by Adam LB