Building tools so intelligence isn't gated by labs, geography, or money.

I am turning ArthaLabs into a focused research program for cheap, reliable, local intelligence: hidden-state readouts for false-pass detection, continual-learning memory systems, and low-resource model training that does not assume frontier-lab budgets.

Most days are split between experiments, benchmark harnesses, careful writeups, and making the public evidence packet cleaner than the claims around it.