HarrisonKim commited on
Commit
d44bf53
·
verified ·
1 Parent(s): b7d6bc3

Update the organization card

Browse files

Describes WIGTN as an AI company with three teams (Solutions, Product, Research) instead of an open-source research crew, lists the model and the dataset with what each is, and links the website, GitHub, WIGTN Tech and contact address.

Files changed (1) hide show
  1. README.md +6 -3
README.md CHANGED
@@ -1,5 +1,8 @@
1
- Hi! We are **WIGTN**, an open-source AI research crew.
2
 
3
- We build practical, domain-specialized AI tools — starting with document intelligence for Korean public sector documents. Our first release is **WigtnOCR**, a VLM-based document parser fine-tuned for Korean government documents.
4
 
5
- Feel free to check out our [website](https://wigtn.com) and try our latest models!
 
 
 
 
1
+ **WIGTN** is an AI company in Korea with three teams. Solutions takes AI transformation (AX) for enterprises from consulting to the systems that run it. Product ships the open-source tools we build and use ourselves. Research publishes peer-reviewed work on LLMs, their evaluation, and speech.
2
 
3
+ On Hugging Face:
4
 
5
+ - **[WigtnOCR-2B](https://huggingface.co/Wigtn/Qwen3-VL-2B-WigtnOCR)**: a 2B document parser for Korean government documents and English papers, distilled from Qwen3-VL-30B into Qwen3-VL-2B with LoRA. It matches or beats its teacher on 3 of 5 OmniDocBench metrics. Training and evaluation code: [wigtnOCR-v1](https://github.com/wigtn/wigtnOCR-v1).
6
+ - **[KoGovDoc-Bench](https://huggingface.co/datasets/Wigtn/KoGovDoc-Bench)**: the 294 validation pages WigtnOCR-2B was evaluated on. The RCPS paper (EMNLP 2026 Industry Track) builds its 663 question–answer probe on the same pages.
7
+
8
+ [wigtn.com](https://wigtn.com) · [GitHub](https://github.com/wigtn) · [WIGTN Tech](https://tech.wigtn.com) · [contact@wigtn.com](mailto:contact@wigtn.com)