Ivan Lok

About me

Ivan Lok

Hi, I’m Ivan. I study Computer Science and General Business Management at HKUST. My interests include computer architecture, efficient AI systems, and embedded robotics.

I’ve spent much of my time at HKUST building robots, leading the Intelligent Racing software team, and teaching new members. More recently, I’ve been working on accelerator design and GPU inference profiling.

Fall 2026 · Carnegie Mellon University
On exchange and conducting undergraduate research with the CAOS Group.

Portrait of Ivan Lok

Education

Sep 2023 – Expected Aug 2028

Hong Kong University of Science and Technology

Technology and Management Dual Degree Program
BEng in Computer Science with an Extended Major in Artificial Intelligence
BBA in General Business Management

4.052 / 4.3Cumulative GPA · Summer 2026

4.27 / 4.3Computer Science GPA

Dean’s List every regular semester on record, Fall 2023 through Spring 2026.

Postgraduate coursework:

Noteworthy coursework:

  • COMP 4211 — Machine Learning (A+)
  • COMP 4421 — Image Processing (A+)
HKUST’s Red Bird sundial at the entrance piazza
Photo: HKUST

Fall 2026 / Undergraduate exchange

Carnegie Mellon University

Current coursework:

  • 10-423 — Generative AI
  • 10-405 — Machine Learning with Large Datasets

Selected honours

  • HSBC Global Scholarship2026–2027
  • ICPC Asia Hong Kong RegionalGold Medal · 11th place · 2024
  • Cathay HackathonFinalist · 2025

Robotics

Jan 2024 – Present

HKUST Robotics Team

Intelligent Racing

HKUST Robotics Team group photograph

Sep 2025 – Present

Senior Software Engineer

Led an approximately 10-member multidisciplinary team developing a drone and tethered ground vehicle. Architected the embedded software stack and implemented cascaded position, attitude, and ESC motor-speed PID control on a custom Infineon platform.

Developed register-level CAN bus and hardware I²C libraries, and advised succeeding team leadership on technical planning and knowledge transfer.

Sep 2024 – Aug 2025

Software Lead & Treasurer

Led a 9-member software team and co-led the 24-member Intelligent Racing sub-team, overseeing technical strategy, cross-functional coordination, and project budgeting. Architected the cross-platform HAL and led the first-year relaunch of the Intelligent Car Race programme.

Jan 2024 – Aug 2024

Software Engineer

Developed embedded software, MCU interfaces, and sensor and actuator integrations, supporting on-robot testing and competition preparation.

Robot Design Competition

Annual HKUST programme · 2024 & 2025

In 2024, I led the software tutorial cycle, delivered embedded-systems lectures to approximately 100 trainees, and mentored teams through implementation and final integration.

In 2025, I led all software activities and co-led the overall programme for approximately 250 participants, covering recruitment, resource allocation, game-rule formulation, teaching, and mentorship.

Selected Projects

Gemmini accelerator architecture with scratchpad, systolic array, and accumulator

Accelerator co-design / Spring 2026

Gemmini Online Attention
& Attention Fusion

I designed a Chisel OnlineAttention module and integrated it with Gemmini’s datapath and RoCC interface, keeping attention weights on-chip and eliminating their O(N²) DRAM round trip.

2.49×attention-core speedup
1.32×full attention-sublayer speedup

At sequence length 128 · cycle-accurate Verilator simulation. Block-wise variants remained stable through sequence length 512.

Chisel / RTL · Gemmini / Chipyard · Bare-metal C

Design, experiments & code
YOLO26n latency comparison across compact, engine-NMS, and dense output routes

Edge AI / Spring 2026

YOLO26n TensorRT
Profiling on Jetson

I compared 15 deployment pipelines and traced a TensorRT version regression with Nsight Systems and Nsight Compute. Output contracts and post-processing can outweigh precision gains.

15controlled deployment pipelines
+1.44 msmatched FP16 version regression

Fastest measured: 7.64 ms/image, with mAP50–95 decreasing from 0.397 to 0.362. Batch 1 · five-run averages over 5,000 COCO val2017 images.

TensorRT · Jetson Orin Nano Super · NSYS / NCU

Findings, benchmarks & code

Embedded systems

Cross-Platform Hardware Abstraction Layer

I architected a reusable C++ hardware abstraction layer across STM32, STC, and Infineon MCUs, separating sensor and control modules from board-specific drivers. Shared code supports Sumo, air-cushion, and drone projects through configuration changes.

Robotics experience ↘

Product engineering

Study Pathway Planner

I redesigned and rebuilt the React front end of an existing degree-planning platform, taking ownership of architecture and UI/UX. Drag-and-drop semester planning helps dual-degree students explore alternative multi-year study pathways.

HKUST DDP Programming Team · Sep 2024 – Jan 2025

Research Experience

Fall 2026

Carnegie Mellon University

Computer Architecture and Operating Systems (CAOS) Group

Undergraduate research

Currently conducting undergraduate research with the CAOS Group while on exchange at CMU.

Jun 2025 – Jan 2026

HKUST

Automated Reasoning and Transformation of Software Research Group

Undergraduate Research Intern

  • Reproduced and redeployed components of an existing LLM-based compiler pass-sequence autotuning pipeline, integrating model workflows with LLVM pass execution and evaluation.
  • Investigated candidate LLVM IR and static-analysis signals for future pass-sequence optimisation.
  • Developed diversity-driven synthetic C/C++ data generation for a WebAssembly-to-C++ decompiler, targeting structures such as function-pointer arrays and indirect dispatch.

Contact

Feel free to email me about research, engineering opportunities, or collaborations.

[email protected]