A service for extracting expense data from receipt images
https://underdoc.io
I built an API for extracting expense data from images (receipt, invoice, demand notes, etc.). I specifically test it for Asian languages (e.g. Chinese, Japanese, etc.)
Would like to connect to those who share the similar interest and be my advisor, mentor.
That's an interesting niche! Curious — did you run into any big challenges working with Asian character sets (like Chinese or Japanese) when extracting data?
Yes, I did some research and tried many open source model like Lamma 3.2 vision, Qwen VL, etc. Found many of them doesn't support Asian language very well.
Found commercial models were performing better.
I posted an article on Medium for my evaluation on gpt-4o and Gemini 2.0 Flash:
https://medium.com/underdoc/llm-evaluation-for-expense-document-image-understanding-d9b3549970b7
Feel free to take a look and let me know if you have any comments. Cheers.