Tuning custom Gemma models for high speed computer vision

In this interview from Google I/O, Matt Rogers, founder and CEO of Mill, breaks down how hardware engineering and cloud architecture are combining to solve the global problem of food waste. Learn how the company deploys hand tuned, custom Gemma models directly at the edge on Nvidia Jetson devices to handle high frame rate computer vision processing (120–240 FPS) over 5 terabytes of labeled data. Discover the system mechanics behind edge-to-cloud data loops, local mass estimation workloads, and how resulting predictive data pipelines can be ingested into enterprise cloud environments to drive agentic procurement workflows for commercial kitchens. Watch more Google I/O Interviews → https://goo.gle/io-tech-chats 🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech #GoogleIO #GoogleCloud Speakers: Arthur Soroken, Matt Rogers Products Mentioned: Gemini