At Ema, I own the multi-cloud deployment platform that provisions customer environments across AWS, GCP, and on-prem. It cut provisioning time from two days to two hours, and now supports production customer environments with cost guardrails and lifecycle automation.
I standardized model deployment on Knative and KServe, moving inference to scale-to-zero serverless services. Alongside workload right-sizing and lifecycle automation, this helped reduce cloud costs by 30%; I also shipped an in-house feature-flag platform that replaced Statsig.
At Ema, I built secure sandboxes for untrusted LLM-generated code and delivered disaster-recovery and supply-chain infrastructure supporting SOC2 and FIPS compliance. Before that, at Fiddler AI and MarkovML, I worked on self-service infrastructure and ML workflow execution platforms.
