Hemlock — launch film.

Julian speaks to camera with the bracelet on his wrist and the robot arm it trains beside him, then the film cuts to skilled workers at a sink, an engine and a workbench, with hand-tracking and force readouts over the footage.

It ends on Hemlock's own result: the same nine minutes of demonstrations, once as video only and once with the bracelet's data, and a tenfold higher success rate. WERZ went from a Thursday scoping call to a wrapped two-camera shoot by Monday, mid-batch.

This is a descriptive summary for accessibility and search discovery, not a verbatim dialogue transcript.

Hi, my name is Julian, and I'm building Hemlock. This is the bracelet. It reads the muscular activity in a user's wrist, so that we can understand how people move their hands and apply force with their fingers. We equip all types of skilled workers with the bracelet, collecting the data needed to understand how they move their hands, apply force with their fingers, and make contact with objects and tools. Essentially, all the information needed to compute how they use their hands to do their jobs well. Information-rich data, zero occlusion, collected at scale. The next generation of robots need to understand real-world force and contact dynamics in order to execute both safely and successfully. We fine-tuned the model twice on a simple egg placement task. Exact same nine minutes of demo data, once containing only video, and once pairing video with the bracelet's data. Success increased tenfold. If you're trailblazing foundation model training, revolutionizing robotic dexterity, or breaking ground in physical AI in any way, please don't be a stranger. Reach out for a chat and some sample data.