One interface. Multiple memory systems.
CPU–GPU ring buffers used by Perception, Behavior, Remote Operations and logging, across onboard and simulation environments.
Systems software · Runtime · Architecture
From the memory model
to the whole machine.
I shape systems architecture across teams and lead execution—from runtime libraries to vehicle integration. At Nuro, that work delivered a new vehicle’s first autonomous miles.
Explore the work ↓ A system, across every layer
NVIDIA DRIVETHORARM64 · CPU / GPU
PCIetransport + synchronization
NVIDIA DRIVETHORARM64 · CPU / GPU
IPC · MEMORY · SCHEDULING · I/O
Dual-compute platform Architecture illustration
Hardware aware.System wide.
Currently Nuro Previously Microsoft Azure
Selected work
Three projects.
A common systems perspective.
Define the right abstractions. Find the limiting mechanism. Make the result work in production.
Inside the platform
The details are
where the system works.
Reusable interfaces above. Careful engineering below. Explore the mechanisms behind the vehicle bring-up.
Technical direction
Set the direction.
Make it real.
A design others could own.
Set the accelerator-first direction for Remote Operations video. A performance engineer made the implementation trade-offs, partnered with the product team and delivered the production transition.
From one vehicle to parallel work.
A week of intensive bring-up produced repeatable startup procedures across fewer than ten vehicles, allowing component teams to investigate their own integration gaps.
This was autonomy-ready startup; autonomous driving followed as a separate milestone.Engineering philosophy
A clear goal.
A simpler path.
Fail loud.
Start with the goal state. Make failures explicit and diagnosable. Keep the design simple—reliability depends on it.
Vision without execution
is hallucination.
Architecture has to survive implementation. Set direction, resolve the hard boundaries, and help others carry it into production.
Iterate fast.
Maximize fun.
Use AI to shorten the loop from idea to working software. Keep engineering judgment in the loop—and make the next iteration worth exploring.
In practice: organization-level design reviews, code-review practices, and production tools that make reasoning explicit.
Experience
Systems, at different scales.
From cloud hosts and autonomous vehicles to connected health devices.
Nuro
2022 — presentDelivered the first Nuro Driver autonomous miles on the Lucid Gravity SUV ↗, running on dual NVIDIA DRIVE Thor compute. The program unlocked strategic investment ↗ from Uber.
Led execution across system integration, platform bring-up, runtime libraries and performance.
Set architecture and roadmap for vehicle-to-cloud logging and offloading. Built production AI-native code review.
Microsoft
2020 — 2021 · 2017Azure Core OS
Host-update orchestration and kernel optimization for availability across millions of Azure hosts, including VM-preserving host updates ↗ highlighted by Azure CTO Mark Russinovich.
Azure Compute & Software Analytics
Cluster capacity planning and VM downtime RCA visualization for Azure. Saved ~$660k/mo via hot-cache and cluster autoscaling policies.
University of Washington
2018 — 2019Health IoT
Built a smartwatch for childhood asthma management and low-power wireless headphones with on-device ML for scene-aware noise cancellation.
Morgan Stanley
Summer 2016Listed Derivatives · Application & backend systems
Hybrid iOS application and Java backend development for listed-derivative products.
BEng, Computer Science & Technology · Xi’an Jiaotong University · 2014–2018