Hacker News Reader

You can create an account and get 50 pages to test for free on your own documents. No training is needed. Any feedback would be greatly appreciated.

View on HN

Pretty impressed. Can you explain how you are getting it done? What are the privacy issues anyone can face?

Hi, sure. We use a combination of OCR, IDP and LLMs, after months of testing on different algorithms and solutions. We make all reasonable efforts to offer great privacy and we don't use customers data for algorithm training. Also, for increased accuracy, we can deploy our solution on the customer own cloud, so the data is store on their environment.

It looks really good. I have tried a PDF and asked for 4 keys and an object, and managed to extract them with high accuracy!

Feedback:

1. Entering fields manually can be hard for long documents; it would be great to upload a file, set boundaries (such as canvas) and define these boundaries.

2. You've mentioned you're GDPR compliant; with this kind of data extraction, you'd mostly want to go SOC2 in the future.

Do you offer a self-hosted version?

Hi,

Thanks for the great feedback. We plan to go on SOC2 in the near future. We can also deploy our solution on customers cloud, so they can own the data.

Reply on news.ycombinator.com