Engineering work
Throughout most of my career I've worked on performance-critical systems. I've used instrumentation to understand system behavior, then use that understanding to change the program, compiler, or runtime so as to improve the measured metrics.
At Meta, I work on the Triton ML compiler and GPU kernel performance for AMD hardware. The work sits at the boundary between compiler engineering and applied performance research, with public contributions visible on GitHub.
I’ve improved algorithms for MI300 systems, implemented a compiler lowering improvement for AMD hardware, and worked on the interface used to express GPU kernels. The focus is HPC-scale performance, numerical accuracy, and architecture-specific behavior.
Independent projects
Outside of formal roles, I’ve hacked on projects that make system behavior explicit. In mach_fun, that meant Mach IPC, exception handling, and runtime image inspection. ald began as an experimental Mach-O linker, while dazisa sketches a VLIW ISA and compiler-scheduling project. This site is generated by hpp, a small Python preprocessor built to keep it static, reusable, and JavaScript-free.
At Adyen, I joined the first U.S. engineering office as the first Staff engineer hired locally in Chicago. The role was primarily technical leadership.
I helped connect engineering work between Amsterdam and Chicago, brought large-scale engineering experience into a new office, and worked to mature the local engineering culture.
At Spotify, I worked on iOS app quality and performance, starting from the premise that performance work depends on measurement quality. I investigated how iOS prewarming changed the meaning of startup measurements and reverse-engineered enough of the underlying behavior to identify practical workarounds.
I also helped design early quality tests that exercised the exact builds shipped to users and collected measurements from outside the app. That setup was meant to make performance and reliability regressions visible before they reached production.
At Snap, I built performance instrumentation and CI infrastructure for iOS at production scale. The most important work was independent performance analysis, including startup-time improvements from targeted I/O optimizations and tooling that helped other teams see regressions before they reached users.
This included local and production instrumentation adopted by dozens of teams, plus a 5–10% startup-time improvement from targeted I/O work.
At Facebook, I worked on mobile performance across iOS and Android. The work mixed measurement, compiler analysis, and systems archaeology. I traced CPU, I/O, startup, and binary-size problems through app code, bytecode, compiler output, operating-system behavior, and AOSP internals, then used those investigations to produce measurable production wins.
That work produced 5–40% improvements in CPU, I/O, and startup behavior across large-scale mobile systems. One representative project was an independent Redex/IODI investigation that found low-entropy binary regions and led to metadata optimizations.
I also contributed LLVM Objective-C++ code-generation fixes (D41050, D59873) and JavaScript startup optimizations in Hermes.
At Arity, I worked on an iOS driving-behavior application used by tens of thousands of users. This role is less central to my current direction, but it was an early place where I used programming-language techniques to make algorithmic experimentation easier in a production mobile app.
At StageBloc, I built and maintained production iOS applications across social, shopping, and SVOD products. This was early mobile product work rather than the performance and systems work that later became my focus.
At Google, I worked on the iOS Google Now feature team, implementing flight and sports cards used by hundreds of thousands of users. It was a brief early role and gave me exposure to large-scale mobile product engineering.