Hacker News

Show HN: Open-source, cross platform document data extraction with no OCR

Hacker News - Fri, 04/11/2025 - 4:02am

The traditional pipeline for unstructured data extraction typically follows these steps: 1. Image → OCR Model (e.g., Google Vision) → Layout Model (e.g. Surya) → LLM → Final Answer However, this can be streamlined using a Vision-Language Model (VLM): 2. Image → VLM → Final Answer

Recently VLMs have improved a lot for OCR and document understanding tasks, specifically the Qwen-2.5-VL series. We can run the Qwen-2.5-VL-7B-AWQ model locally with just 16GB VRAM, and perform end-to-end information extraction (fields and table extraction) without any external models.

Hallucination with VLMs One question I am asked often is, higher hallucination with VLMs compared to OCR model. This is a valid point. But, even with correct OCR and Layout formatting, LLM can still hallucinate, and can give incorrect final answers. Layout models often struggle with complex documents (e.g., tables, complex sparse document). If the formatted text from the layout model is incorrect the LLM model will always produce incorrect extraction with high confidence.

Check out our GitHub repo for implementation details:

GitHub: https://github.com/NanoNets/docext

Comments URL: https://news.ycombinator.com/item?id=43651477

Points: 2

# Comments: 0

Categories: Hacker News

Show HN: Basic RESTful boilerplate – Express with TypeScript and hot-reloading

Hacker News - Fri, 04/11/2025 - 4:01am

A simple RESTful boilerplate made with Express that supports TypeScript and has hot-reloading.

Comments URL: https://news.ycombinator.com/item?id=43651471

Points: 1

# Comments: 0

Categories: Hacker News

Wernher von Braun

Hacker News - Fri, 04/11/2025 - 3:35am
Categories: Hacker News

Show HN: Fanagrams – The (Fake) Anagram Challenge

Hacker News - Fri, 04/11/2025 - 3:32am

I just launched Fanagrams, a fast-paced challenge that puts your word skills to the test.

Every round serves up a real word, and your mission is to sift through a handful of (fake) anagrams to pick the correct one. But here's the twist: you only have as many seconds as the word has letters!

If you’re into quick puzzles, wordplay, or just love a challenge that keeps you on your toes, give it a try!

Would love to hear your thoughts and ideas for making it even better. Happy (fake) anagramming!

https://fanagrams.vercel.app/

Comments URL: https://news.ycombinator.com/item?id=43651355

Points: 1

# Comments: 0

Categories: Hacker News

MeshCore

Hacker News - Fri, 04/11/2025 - 3:25am

Article URL: https://meshcore.co.uk/

Comments URL: https://news.ycombinator.com/item?id=43651333

Points: 1

# Comments: 0

Categories: Hacker News

Pages