Software Engineer III · Google | Stanford AI (Graduate Certificate)
I build systems that run at scale. Currently on Google Search TPU Efficiency — optimizing ML training workloads across Google's TPU fleet and shipping infrastructure that has saved 100+ SWE-years of compute to date. Previously on Search Infrastructure, productionizing the AI services behind AI Overviews and AI Mode at 100K+ QPS.
I care deeply about observability, reliability, and the intersection of systems engineering with machine learning. Outside of work I'm pursuing a graduate certificate in AI at Stanford, and I enjoy open-source work, climbing, and anything outdoors.
Software Engineer III · Search TPU Efficiency
Improving runtime efficiency of ML training workloads running on Google's TPU fleet. Focus on infrastructure optimization, observability, and policy enforcement to maximize resource utilization. Released a handful of optimizations accounting for 100+ SWE-years of savings to date.
Software Engineer II/III · Search Infrastructure
Software Reliability Engineer Fellowship
Intensive SRE training covering Linux systems, scripting, systems design, databases, CI/CD, monitoring, and troubleshooting at scale.
Software Engineer Intern · Cloud Run, Search Fulfillment, Search Web Architecture
Three internships across Google teams: Kubernetes containers on Cloud Run, assembling responses for Search Fulfillment, and frontend infrastructure for Search Web Architecture.
Software Engineer Intern · Rentals (×2)
Worked on pricing algorithms for Lyft Rentals, introducing profile-based pricing branches as the company expanded into Lyft Rentals for Businesses.
Software Engineer Intern · Fresenius Medical Care
Wrote full test suites (unit, functional, DiRT, E2E) for a medical equipment lifecycle management system. Client purchased the project for $10MM.
Graduate Certificate in Artificial Intelligence
B.S. in Computer Science · 94/100
My open-source work lives at github.com/dmosc. A selection of highlights below — more coming as I find time outside of work.
Explore all public repositories, tools, and experiments across systems programming, ML, and web development.
My research interests sit at the intersection of ML systems, infrastructure efficiency, and large-scale distributed systems. Through the Stanford AI graduate program I'm deepening foundations in machine learning with a focus on applying them to production systems challenges.
LeWorldModel Embeddings are Misspecified for Control
Evaluates whether JEPA-style world model embeddings support long-horizon planning on OGBench PointMaze and Scene. Finds planning degrades monotonically with start–goal offset, traces the failure to the representation rather than the dynamics or planning budget, and argues the task-agnostic predictive objective is fundamentally misspecified for control.
Read paper →Predicting TikTok Virality with Late-Fusion Multimodal Learning
Develops a multimodal ML framework fusing video, text, and tabular metadata to estimate the probability a TikTok video goes viral, combining intrinsic content signals with extrinsic creator/account signals over a curated TikTok-10M-derived dataset.
Read paper →