Martin Nguyen
M.S. Computer Science, University of Illinois at Chicago
Software engineer · GPU & systems performance
About
I'm a software engineer with a system-oriented mindset and a focus on performance. My work sits at the intersection of low-level systems, GPU programming, and machine learning — building runtimes, kernels, and event-driven pipelines that hold up under real workloads.
Before UIC I was a founding engineer at Vizgard in London, where I led the design of a real-time computer vision pipeline serving dozens of HD camera streams on a single box, and earlier did research engineering at VinAI Research (since acquired by Qualcomm) on efficient face analytics models.
I'm currently a graduate student at UIC, working on Transformer inference and GPU memory consistency.
Research & Engineering Interests
- High-performance GPU inference. Custom CUDA / Triton kernels, paged attention, weak memory models, and the boring-but-essential work of profiling traces until they tell you the truth.
- Real-time ML systems. Lock-free pipelines, event-driven architectures, and the latency ceiling that separates a demo from a deployed product.
- Kernel-level networking. eBPF / XDP for in-kernel packet processing and the kinds of policies that are too hot for userspace.
- Generative & multimodal models. LLM / VLM serving, distillation, quantization — making large models fit where they otherwise wouldn't.
Selected Work
all projects →Black Hole Raymarcher
WebGPU · WGSL · 2025 →A real-time gravitational-lensing renderer that integrates the Schwarzschild geodesic ODE per pixel — entirely in a WebGPU fragment shader, with a procedural accretion disk and HDR star map.
Weak Memory Behavior on GPUs
PTX · CUDA · 2025 ↗An empirical study of NVIDIA's weak memory model — characterizing the synchronization primitives needed for correctness under contention, and the performance cost of getting them right.
Dynamic eBPF Firewall for DDoS Mitigation
Rust · eBPF / XDP · 2025 ↗An XDP-attached firewall written in Rust with hot-reloadable rules, used to study trade-offs between in-kernel filtering and userspace policy engines.
Recent Writing
all posts →Contact
Best reached by email at manhntm3@gmail.com. For collaboration, code, or research conversations — open issues or DMs are welcome.