On Mashable: Building AI Infrastructure at Billion-User Scale
Mashable profiled my work leading AI infrastructure at billion-user scale: lifting model deployment success from 60% to 99%, cutting scale-up time from three days to two hours, and a GPU efficiency redesign worth an estimated $20M in savings.
Mashable featured my work on AI infrastructure and large-scale model deployment, profiling how we made AI deployment reliable at a scale of more than a billion users.
The piece covers the "AI operating model" framework I led to fix model deployment reliability — taking deployment success from around 60% to nearly 99% across nearly 4,000 deployments, cutting model scale-up time from three days to two hours, and getting new post-training experiments into production traffic in under 30 minutes. It also covers a GPU efficiency redesign on one of our advertising models that increased throughput 4.4x per GPU (roughly 90 to 400 queries per second) without adding latency — worth an estimated $20 million in GPU capital savings. And it traces the infrastructure work before AI became the industry's biggest story: a unified hardware forecasting platform, an enterprise-wide reliability transformation across 178 engineering and 47 SRE teams, and one of the largest enterprise cloud migrations in retail.
