I moved to Data Infrastructure initially to stabilize and up-level the area as its first staff engineer following a period of significant talent loss. Initially I led systematic investments in reliability engineering and on-call tuning, stabilizing and automating releases of our legacy Hive/YARN on AWS stack, which paved the way to pilot a major overhaul and stopped attrition for the year. Later, I kept the area focused on safe and within-budget migration to Databricks by driving strategy for a variety of technology upgrades from major MySQL version updates to systematic platform management. Throughout this time period I also strengthened our management practices and mentored dozens of engineers ranging from talented new hires learning about direction-setting to mid-career and senior engineers working towards and successfully reaching staff engineer.

I moved to Paper Infrastructure where I have led several projects addressing key gaps in feature support, standards adoption, and performance. I led development of a new architecture to safely serve arbitrary Dropbox functionality in our staging environment with access to production packaging/maintenance patterns and real AWS resources, which unblocked feature development for dozens of product and infrastructure engineers. I led performance tuning efforts that dramatically lowered server-side p95 latencies for many handlers, including doc creation (50%) and rendering (25%), adopting industry-standard interfaces and closing release-testing gaps. I led consolidation of heterogeneous developer environments to Dropbox’s standard development instances and lowered the memory usage of the company’s base product development container by half, normalizing our deduplication strategy and reducing development costs. I led conversion of Paper’s monolith server-side application from running nakedly on manually-allocated, provisioned-out-of-band hosts to running in Dropbox’s standard container environment, automating maintenance and modernizing instance types.

I led Fleet Management, a team focused on production resilience that owns Dropbox's services and tooling for host allocation, maintenance automation, and reliability testing. The team worked closely with other teams cross-functionally to ensure that resources are smoothly circulated into and out of active use, and that services are tested regularly against reliability standards. I brought on a dedicated tech lead to focus on managing the team in the second half of 2018, during which time we expanded Dropbox's capabilities to accurately model and plan physical infrastructure spend, consolidated the infrastructure services that manage the full lifecycle of hosts as they circulate between business needs, provided automated AWS instance snapshot refreshes, and built-in support for emerging transparent-encryption-at-rest standards. Systems that were developed by junior engineers on my team during this time are still working well as of 2024!

Jul 2017 - Jul 2018

During this period I was the tech lead for the new Fleet Management team I launched to own several facets of production resilience at Dropbox, including services and tooling for host allocation, maintenance automation, and reliability testing. We worked closely with other teams cross-functionally to ensure that resources are smoothly circulated into and out of active use and that services are tested against reliability standards.

Jan 2015 - Jun 2017

During this period I was a senior engineer on the team tasked with Dropbox's server-side build and push infrastructure. I was responsible for unifying disparate push models and rewriting the associated services backwards-compatibly from scratch. I also contributed several key reliability-enhancing features including push previews and a built-in canary process. In parallel, we also overhauled the standard runtime environment in use at Dropbox to provide effective service management to the majority of Dropbox engineers. The team has also operated a machine-inventory service, the production-permissions service, and service-discovery infrastructure, and continues to play a key role in supporting our use of PoP and EC2 regions around the globe.

As a software engineer at D.E. Shaw Research (DESRES), I was active in improving analysis infrastructure, experimenting with various data formats, clarifying access patterns, and optimizing hardware performance. I also characterized the errors introduced by varying the precision of a simulation's coordinates and implemented algorithms for enhanced sampling of the free-energy landscape during simulations.

In my earlier time at DESRES, I worked on a variety of biophysical modeling projects. A particular focus was heterotrimeric G proteins, a family of peripheral membrane proteins that is responsible for transducing a constellation of signals in cell biology, and the upstream G-protein coupled receptors that come up ubiquitously as drug targets in modern medicine. The majority of this work, which bore several peer-reviewed publications in Science, Nature, and other journals, was reading scientific literature, kicking off simulations based on atomic-resolution crystal structures, and comparing the two.