unstract VS mitta-community

Fast changing libraries are a huge pain. That's why a no-code approach like Unstract (https://github.com/zipstack/unstract) makes sense.

Hey HN!

Over the last few months, we’ve seen many tools here trying to tackle the problem of making complex, unstructured documents ready for LLMs.

The complexity primarily includes parsing highly complex documents in terms of format, layout, design, complex tables, checkboxes, etc, with high accuracy and reliability.

LLMWhisperer is our take on the problem. LLMwhisperer solves most of the document complexity with reliable accuracy.

With our user-friendly playground (https://pg.llmwhisperer.unstract.com/), you can effortlessly test your document use case. No sign-up is required. We eagerly await your valuable feedback.

LLMWhisperer is also a part of Unstract, an open-source document processing automation platform.

Links: https://llmwhisperer.unstract.com/

Playground: https://pg.llmwhisperer.unstract.com/

Unstract: https://github.com/Zipstack/unstract

Feature screenshot: https://imgur.com/a/QtrqNbB

If anyone is interested in exploring this space, try another similar tool LLMWhisperer (https://llmwhisperer.unstract.com/). It is a part of Unstract, an open-source document processing tool (https://github.com/Zipstack/unstract)

Unstract supports a variety of providers for LLMs, Vector Databases, Embeddings, Cloud File Storage systems and databases/data warehouses. A full list is available on our Github page (https://github.com/Zipstack/unstract).

No-code Prompt Studio:

Here's an EasyOCR service: https://github.com/MittaAI/mitta-community/tree/main/service.... A PDF to image processor is being built and should be out in a few weeks.

No docs, but happy to help anyone wanting to use it. Email is kord @ the company I'm working on.

I built this with Playwright and OpenAI's function calling stuff (sorry, no time for docs): https://github.com/MittaAI/mitta-community/tree/main/service...

My thought was to put the results of this in a vector store, with any errors that resulted as opposed to wasting time training a model.

Source for the FFmpeg build (no AI): https://github.com/MittaAI/mitta-community/tree/main/service...

I just built a translate pipeline for MittaAI: https://github.com/MittaAI/mitta-community/tree/main/cookboo.... The pipelines crawl and translate any publicly available page. The `translate` pipeline uses Gemini, but could be changed to another model for use.

If you want a generalized version, try this: https://github.com/MittaAI/mitta-community/tree/main/cookboo...

The query pipeline isn't that sophisticated, but it could be altered to do page reference and use keyterms first to filter, instead of doing the vector similarity on all data.

> A service to preprocess the data with custom prompts would be neat.

It is pretty cool to mess about with it. I posted it last week, but didn't get any nibbles: https://github.com/MittaAI/mitta-community/tree/main/cookboo...

It doesn't matter if the demos were good or not. Demos don't always go as planned. I've found using Gemini for image scene description and chat works well, and have it running a portion of an image scene to spoken "thoughts" pipeline using ElevenLabs for the voice: https://github.com/MittaAI/mitta-community/tree/main/cookboo...