DingusCorp Labs

"thats some sick beat"

i'm not a critic or professional i just like to have fun :D

Open Notebook Test Suite

I've found myself using Open Notebook a lot for school and the like. As such, I tend to collect a LOT of sources. Usually over 40, sometimes 70 per-subject. Open Notebook can't always handle that, especially when running on limited resources.

So just to figure out what any given instance can handle, I put together a small, highly configurable utility to generate hundreds of fake sources to upload to Open Notebook and test the capacity of your setup.

How it does it

It simulates a wide variety of sources: Audio (STT), Images (OCR), word documents (Docx), slideshows, PDFs, and HTML for websites. Depending on the content type and configuration, it will either SIMULATE content processing errors, or attempt to INDUCE them. For audio: You can optionally add a TTS endpoint. The LLM will generate text for the source and then turn it into speech then overlay sounds and noise over it from a folder to simulate a real recording. If you do not attach a TTS endpoint, it swaps to SIMULATION mode and instead the LLM just spits out Markdown and tries its best to emulate the transcription errors a STT model might make. For OCR: It's somewhat similar? It just tries to emulate the errors an OCR engine might make. For Docx: It generates real Docx files with complex graphics and tables. For PPTX: The same as Docx. It generates a real slideshow with complex tables and graphics. For HTML: Generates real html- BUT it's INVALID HTML. It's still valid enough to be correctly rendered in a web browser, but it's technically incorrect. For testing edge cases with document content extraction For PDF: Same as Docx You can configure the topics for all of these by providing a title and prompt in the config, so you can tailor this to your actual open notebook use. It's written in Node.JS (of course) and supports Parrallel Inference if you have it (And you can optionally set a max for the amount of concurent jobs).
You can also configure the exact features of each source type, how many sources there are, or use PSRNG to add a level of varience to the amount of sources. Currently, the program just outputs all the sources to folders that you can batch upload. I'm planning on implementing the Open Notebook API directly in the future so that the thing can be configured to upload automatically.

Disclaimer

A good chunk of this project was written using AI. Specifically, using the OpenCode harness and a mix of Qwen 3.5 9b and Gemma 4 12b QAT.

Q: Why AI? Why would you vibecode something??

A: It's a one-off utility thing I needed for my own purposes. I'll be putting it up on the warehouse once it's done, but as it's not a major project of mine, I used AI to speed it along. It may be more prone to bugs and vulnrabilities.

I don't use AI in any major portion on my passion projects. A very small portion of DoofOnline (this site) was AI assited, and some earlier components are strung together from Stack Overflow, but more often then I ended up spending more time correcting the LLM's mistakes then I would've if I had just done it myself.
Plus, this site is MY passion project. It's something I enjoy doing otherwise I wouldn't be doing it. It breaks the site's philosohpy and ruins my fun.