Kodifly Limited
Role progression
Computer Vision Engineer (L2)Apr 2026 – Present
Localization & estimation
- Re-architected core localization from SLAM to GNSS + IMU + EKF.
- Cut steady-state edge CPU load from ~100% to ~30%.
- Implemented factor-graph optimization (FGO) to fuse inertial trajectories with GPS.
- Localization accuracy improved by ~43% vs the prior fusion stack.
Fleet reliability & tooling
- Deployed a hardened bastion host on GCP.
- Acts as the controlled jump point for remote access across the edge fleet.
- Built an internal triage / remediation platform end to end.
- Prometheus + Grafana for signals and dashboards.
- Next.js front end so incidents are actionable from one place.
Computer Vision Engineer (L1)Nov 2025 – Apr 2026
Imaging pipeline
- Migrated the high-throughput imaging path to CUDA + NVJPEG.
- Frame latency dropped from ~130 ms to ~30 ms (~4×).
Hardware sync & release quality
- Designed and brought up a hardware synchronization PCB.
- Nanosecond-level timestamp alignment between cameras and IMU.
- Introduced mandatory automated testing for the imaging + localization stack.
- Blocked releases that would have failed in production.
Junior AI EngineerJun 2025 – Nov 2025
Bridging lab and deployed hardware
- Curated odd field captures from deployments and cleaned label drift before retraining cycles.
- Debugged ROS + image ingestion paths when edge boxes returned inconsistent metadata or corrupt frames.
- Kept training configs and what shipped on-device aligned so regressions were traceable.
PythonPyTorchCUDAROSDockerGCPPrometheusGrafanaNext.js
Standard Chartered Bank
FinTech Analyst (SME Banking)Jul 2024 - Aug 2024
Delivery & product impact
- Owned feature rollout on the SME Express client journey.
- Monthly active users on that journey rose 15.72% during the engagement.
- Specified SME financing dashboard views.
- Post-launch satisfaction on the new dashboards landed around 83%.
Testing & automation
- Ran UAT and wired results into Azure DevOps.
- CI/CD-friendly pipeline; full regression dropped from an all-day manual pass to under an hour.
- Built automated harnesses for digital payment partnerships.
- Caught defects earlier—roughly a third fewer issues reaching later stages.
- Documented requirements and automated UAT for HKMA’s Commercial Data Interchange onboarding.
- Teams could digitize paper-heavy steps without breaking compliance checks.
Azure DevOpsMicrosoft AzureUAT automationData visualization
The University of Hong Kong
Student RA (Robotics)Jan 2024 - May 2024
- Calibrated a manipulator with vision so point clouds projected into the lab’s digital twin landed within 98.3% of the measured ground truth.
- Ran inverse kinematics and motion planning in MoveIt for the same arm, tightening trajectories before they hit hardware.
- Hooked up vision–language models for zero-shot object classification so new objects could be labeled in the scene without retraining a bespoke classifier each time.
ROSMoveItPythonOpenCVIntel RealSense
Kodifly Limited
Software Engineer (Intern)May 2023 - Aug 2023
ROS + web visualization
- Shipped a browser LiDAR + camera viewer (Vite, Three.js) subscribed to ROS topics.
- Vehicle / pedestrian bounding boxes.
- User-drawn collision volumes to flag traffic through regions of interest.
Synthetic data
- Extended Blender with a LiDAR sampling rig for terrain and hazard scenes.
- Exported synthetic scans so detectors could train without waiting on rare real-world landslide captures.
Vite.JSROSBlenderPythonThree.JS
The University of Hong Kong
Student RAMay 2022 - Apr 2023
VR twin & scale
- Built the Innovation Wing VR twin in Unity for the live Innoshow.
- Hosted roughly 4,000 concurrent visitors during the online event.
Release workflow
- Automated Quest builds with bash.
- Rebuilt Android APKs and pushed installs when headsets connected—no manual sideload before each demo.
Interaction design
- Designed the 3D spatial UI so first-time VR users could navigate exhibits without getting lost in menus.
UnityC#BlenderQuest SDKBash