Computer Engineering senior. I work at the layer where hardware stops being an abstraction — GPU memory, kernel launches, cache behaviour, and latency budgets you can't miss.
- Founder of ERK Security — security & AI tooling: LLM-backed threat analysis, GPU inference pipelines, and a reflexive P4 data plane with an ML control plane.
- Building GPUFlow — live visualization of what every process is actually doing on your GPU. C++17, passive NVML sampling, dependency-free SSE server, zero-install browser UI.
- Edge perception — anomaly detection on Jetson AGX Orin + ZED 2i for UAV workloads.
- Going deep on — CUDA memory hierarchy, TensorRT optimization, multi-target tracking, modern C++ concurrency.
Systems C++17/20 · CUDA · NVML · pybind11 · CMake · Linux · Assembly (MIPS/PIC)
Inference TensorRT · Jetson (AGX Orin) · ONNX · PyTorch
Perception Kalman filtering · multi-target tracking · signal processing · ZED SDK
Security P4/bmv2 · Mininet · adversarial ML · network asset discovery
|
Right now Tiled SGEMM on Ada, chasing the gap between 8.4 and 11.6 TFLOP/s. Shared-memory bank conflicts, register pressure, and the eternal question of whether the bottleneck is the kernel or the copy. Spoiler: it is almost always the copy. |
