Roles where I've shipped research prototypes, production systems, and internal tools across robotics, IoT, and
backend services. Authorized to work in the United States.
Mountain View, CA · July 2026 – Present · Full-time
Help design the end-to-end pipeline for Unitree G1 and Go2: teleop data collection, VLA training, and inference on Jetson Thor/AGX. Turning that into a platform so more of the team can run the loop without each person having to know every piece.
Care about real-time latency in teleop and on-robot inference. Keep the stack current with robotics research and industry practice (Isaac GR00T N1.7, Cosmos Reason 2, Cosmos 3).
Data infrastructure: synchronized camera and sensor streams as MCAP/zstd with Parquet metadata, automated S3 upload, into VLA and RL training.
Hardware across G1, Go2, Jetson Thor/AGX/Nano, Inspire and BrainCo hands, Pico XR, AIRBOT, UR6e, and Velodyne.
Isaac GR00T N1.7Cosmos Reason 2Cosmos 3Ï€0Isaac TeleopGear SonicJetson ThorROS 2
Designed robot learning data infrastructure from scratch for humanoid and quadruped platforms: multimodal capture, chunked S3 upload, and annotation schema for the team to build on.
Built XR teleoperation with NVIDIA Isaac Teleop across Unitree G1, Go2, and Inspire hands. Compared ZMQ, WebRTC, and gRPC for latency.
Built MoRAL on NVIDIA Cosmos Reason 2: first the model learns to read a bird's-eye view we render from LiDAR and radar, then it learns driving reasoning (what is approaching, whether to brake). Presented at IEEE IMC 2026 in Fukuoka. arXiv:2608.02449.
Built and deployed an AMR autonomy stack (MOLA, RTAB-Map, VIO, LiDAR SLAM). Cut a 1,100ms bottleneck with NVIDIA Warp GPU kernels so obstacle avoidance held up on the real robot.
Integrated PointPillars 3D detection with LiDAR and camera fusion into a ROS2 MoveIt navigation stack, including NVIDIA Isaac ROS on Jetson.
Built a collaborative perception pipeline merging two LiDAR streams using ICP, NDT, and RANSAC, achieving 88-92% point cloud overlap.
Managed a 50+ Jetson Orin/Nano fleet via Docker and SSH automation, achieving 90% fleet availability and reducing manual intervention by 40%.
Built teleop pipelines and YouTube tutorials enabling students to replicate the full AMR autonomy stack. Mentored lab members on ROS2, SLAM, and sensor fusion.
Led a team of 6, mentoring 4 junior engineers, managing a physical security platform across 10K+ cameras (AXIS, Avigilon, Hanwha, ONVIF) and access control systems. Primary technical contact for 3 enterprise clients across 15K+ devices, contributing to 35% revenue growth.
Trained and deployed custom YOLO object detection models on live security camera feeds, owning the full CV training pipeline from data collection to production inference. Built an async Python IoT monitoring system with 99% accuracy, reducing false downtime alerts by 87%.
Redesigned the OTA firmware upgrade pipeline using async Python and multithreading, cutting rollout time for 1K+ devices from 5 hours to 7 minutes (97% reduction). Built Python and Golang SDKs adopted by 4K+ IoT devices.
Built a 10K-client IoT simulator using Docker to stress-test cloud-edge behavior before physical deployment, catching 25+ critical race conditions.
Built a B2B SaaS application called Byonic from scratch end to end, praised by customers during PoC. Worked closely with the product team using pair programming to ship features on time.
Implemented REST APIs handling over 10 million records. Migrated from raw SQL to ORM. Deployed to UAT with zero downtime. Built secure multi-tenant auth using JWT tokens.
