We turned cortical tricks into tiny (~1M param) adapters that plug into various neural network architecture models, CLIP, GPT, CNN, etc..
It evolutionarily navigates astronomical searche space of configs in ~1,000 cheap experiments and delivers noticeable (i.e. +8%) accuracy.
No retraining the backbone. No crazy compute. Pure neuroscience cheating codes.
Pay only if it beats your baseline—no improvement, no payment.
Would love brutal feedback from the HN ML crowd — especially if anyone wants to throw their weird VLM at it and see what the brain stuff does
I am also curious to hear what is 'significant' amount? Is it something which a couple of guys can't finance by themselves in a garage? For a month or for a year? We've built one SaaS business without, what I would consider, significant funding.
