Catherine Zhao is co-founder and CEO of Pelica Health, building the AI operating system for value-based care.
Catherine holds a Bachelor's degree from Dartmouth College, where she studied Computer Science, completed the Tuck Business Bridge Program, and spent two years as a memory-care volunteer at Kendal at Hanover. After graduating, she continued clinical work, volunteering as a medical assistant at the San Francisco Free Clinic. Her clinical AI research at Stanford School of Medicine, ongoing since 2014, focuses on deploying AI in clinical workflows in low-resource settings.
At Google, Catherine led billion-user products across YouTube, ChromeOS, and multiple AI efforts. She started as a software engineer at YouTube, where she launched the YouTube Shopping product, the first native shopping experience on the platform. She went on to ship products in ChromeOS, then moved into Google's GenAI organization, where she contributed to work on Gemini and to image and video models including Imagen and Veo. She also built the first enterprise AI agents at Google, deployed to procurement and finance teams. She entered Google through the Associate Product Manager program, started by Marissa Mayer and admitting less than 1% of applicants, as 1 of 50 in her cohort.
She is a member of Women Business Leaders of the US Health Care Industry Foundation and a current participant in Y Combinator (YC X25), which accepts approximately 1% of applicants.
At Pelica, Catherine writes about why operations, not analytics, is the bottleneck in value-based care.
A dashboard tells you what to do. An AI agent does it. Buy execution, not analytics.
The step-by-step anatomy of an autonomous medication-adherence gap closure.
The license fees are the smallest line. The real costs are duplicate outreach and retrospective catch-up.
A founder letter on why we left Google to build the operating system for value-based care.
The three triple-weighted Part D adherence measures move Stars more than any other lever.
NCQA retires hybrid HEDIS by 2029. Supplemental data volumes increase 35x to 75x per measure.
30 minutes. We bring a working canonical record stitched from your sample feeds. You bring the operations problem.