What I'm working on right now. Filter by topic below.
One place to record what was made, how, and what it measured — usable later as training data for machine learning.
Say "hey brain" out loud, and a physically separate Pi runs the arm through a full autonomous pick-and-place — fully local, no cloud anywhere in the chain.
Stage 1's goal is achieved and closed. Three parallel threads now: a wheeled base (LeKiwi), a real language-conditioned policy (SmolVLA), and voice polishing.
Automate playlist creation for a music startup using unsupervised clustering. Several real bugs and one finding that overturned its own headline conclusion mid-analysis.
From a box of parts to fully autonomous pick-and-place — imitation learning with LeRobot, ACT, and a hard-won two-camera setup.
A text-trigger backdoor attempt revealed an architectural dead end — then a corrected, genuinely working visual backdoor confirmed the fix.