I build developer platforms for very large monorepos — Git as a service, merge queues, code review, and CI. My work sits between a reviewed change and a green trunk: speculative execution, queue sharding, and making test signal trustworthy at a scale where "just rerun it" is not an answer.

Paper: GitFarm: Git as a Service for Large-Scale Monorepos — first author. How Uber serves Git for a monorepo past the point where off-the-shelf hosting works.

Uber's speculative merge queue, open sourced — it keeps trunk green at scale by building batches before they land. I'm the top contributor by commits, at roughly 4x the next contributor.

I work across the speculation engine (what to build, in what order, and when evidence is enough to land), the message queue that has to stay correct as tenants grow, and the Git/CI plumbing that makes a speculative batch actually merge. Design docs and RFCs sit alongside the code.

Computes the changed targets and conflict graph for a given base and head. SubmitQueue uses it to decide what to speculate. I helped open-source the project: GitHub Actions, CODEOWNERS, and a required-checks merge gate.

Also: tobi/walgit — wrote the dev setup guide for this Rust project · uber/arcanist — drove the PHP 8.2 upgrade of Uber's fork · sq-demo — burst-traffic demo harness for SubmitQueue · gists — design notes and counter-proposals

_{Open to conversations about developer platforms, build and merge infrastructure, and large-scale CI.}