current

last updated

curious about

what transfers when a policy moves to an environment it hasn't seen, and whether reliability comes from better policies or from the system built around them

building

corrective rl environments for robots, task routers within action policies, and turning failure into training data

reading

revisiting the mom test by rob fitzpatrick. read it for the first time in uni when i learned enterprise sales at my first startup stint

training for

delhi half marathon on october 18th. pushing weekly mileage from 15-20k to 25-30k