I was in openrouter’s website comparing models and there was a section where the output of each model was being tested. In the output text, the first phrase was “Attention is all you need”. I’ve heard that before, I’ve heard that a lot actually but I honestly didn’t know what it was. So off I got to old friend Google searching for this phrase and learn that is a research paper about the core architecture of machine learning (or something like that), quite interesting stuff. Then, on the wikipedia page for this research paper I see this paragraph “… At the time, the focus of the research was on improving Seq2seq techniques for machine translation, but the authors go further in the paper, foreseeing the technique’s potential for other tasks like question answering and what is now known as multimodal generative AI”. What I gathered in the moment was “so these scientist from Google were trying to probably improve Google Translate and in the process end up creating the foundation for how LLMs work today?” (I might be wrong on that assumption but it was at least the thought that gave me the itch).

So then I thought, here is again that phenomenon where a person or a team starts working on a project to solve a problem the currently have in some area. They develop a solution that fixes that problem they have, but then, themselves or someone else has a light bulb idea moment when they realize that the solution the concocted doesn’t only fix the problem they currently have, but it can also be applied to other scenarios that have a lot more potential for impact, and when this happens, that realization moment, becomes the backstory of the great product/solution/invention that became very successful.

So then I thought, do I have anything in my projects drawer might be sitting in the moment RIGHT before that light bulb realization? and the answer came rather quickly, ‘compas’ was the only shy kid playing with a stick in the sand while holding unknown potential.

Now with the spark of curiosity, I need to find out what that potential could be. I try thinking and I come up with nothing, it’s 1:30 AM and hungry so no surprise there, but apparently I had enough brain goo for one more idea. I open pi (for minimal harness intervention) and I query the strongest model I have available and I ask it about this. I give it the core idea of what i built compas for (to give AI agents a search shortcut so they’re not bloating their context during exploration tasks reading irrelevant files when searching for something specific in a repo). It identified this project’s main issue, that every IDE and their mother eventually will release an integrate version of this index+semantic search thingy. But what I did not expect it to do, was to actually come up with real valuable scenarios where compas as a core concept, can really be impactful outside of codebases.

Areas with potential high impact could be:
- Legal documents and contracts
- Healthcare
- Incident logs

A bunch of areas that I have no knowledge of how they work or what they currently use.

So now this inquiry has opened a can of worms and I don’t know where to go from here.

Dig deeper into one of these industries? Turn compas into a core and build a document search solution powered by it?

I don’t know. What I do know is that it’s currently 2:14 am and I need to sleep.