Work
Systems where latency is a correctness property rather than a nice-to-have — execution and trading infrastructure, and simulation infrastructure before that.
Currently working somewhere I am not listing publicly. The history below runs to September 2025.
Execution team — the systems that turn a research signal into orders on a venue, and the latency between the two.
- Removed critical sections from the execution hot path using lock-free algorithms specialised to the underlying hardware.
- Introduced a user-space allocator model built on object pools, cutting allocation cost on the hot path.
- Brought C++20 into the core library — ranges, consteval, and concepts to replace the SFINAE machinery they displace.
- Cut combined unit and integration test time by eight minutes, which is the kind of change that compounds across every engineer on the team.
- Scoped and removed redundant components as part of a larger execution framework migration.
Low-latency trading systems for options market making — market data ingest, order entry, and the execution path in between.
- Built and maintained the systems on the critical path between exchange market data and order entry, where the latency budget is measured in nanoseconds and allocation on the hot path is a bug.
- Worked directly with quantitative researchers to take strategy from idea to production, and to keep the feedback loop between them short.
- Wrote allocation-free replacements for standard-library containers whose behaviour under contention did not suit the workload.
- Removed explicit synchronisation from a producer/consumer path by relying on cache-coherence behaviour specific to Skylake-family cores — the most fun I have had reading errata.
- Built high-fidelity exchange simulators so strategies could be back-tested and parameter-searched against realistic venue behaviour.
Simulation infrastructure for autonomous driving — the systems that let the fleet be tested in software rather than on the road.
- Designed a distributed cache for vehicle perception data that eliminated redundant recomputation across simulation runs, saving an estimated $10M.
- Scaled the service to support two billion simulations, which unlocked training workflows that were previously not runnable at all.
- Cut simulation scheduling time by 80% and end-to-end job latency by 50% at comparable simulation quality.
Search over the AWS Organizations administrative surface.
- Added full-text search filtering to the AWS Groups web UI.
- Wrote the design document and presented it to the wider AWS organisation.
Optimisation research and data visualisation.
- Cut program runtime roughly 400× — about 200 seconds to half a second — by migrating a Python application to C++11.
Client work
Engagements taken independently.
UI and backend for an HR tool built around decentralised identifiers, plus the CI/CD pipeline that shipped it.
Hotel deals site for Spanish-speaking travellers. Took page load time under two seconds by resizing images, moving to WebP, and fixing the caching policy — the same class of work as the front-end performance problems I still find interesting.
Backend for a fantasy sports betting platform settled in ERC20 tokens.
Education
Honours BS, Computer Science
The University of Texas at Dallas · 2017 – 2020
GPA 3.96 / 4.0