Roman Chemet

Roman Chemet (Cactus): On-Device Inference, Hybrid AI, and Ambient Computing

Cactus founder Roman Chemet explains the YC-backed low-latency inference engine for running AI privately on phones, wearables, and laptops — "in a basement and on a plane." He describes YC's rapid 10-minute interview process, argues voice and dictation are the stickiest AI applications, and lays out the "hybrid AI" thesis of local and cloud models handing off tasks, recently stress-tested in a hackathon with DeepMind's Gemma models. They also debate always-on ambient AI wearables and how Europe's privacy norms versus San Francisco's convenience-first culture will shape adoption.

1:44:12