I'm a Master's student in Computer Science at UC San Diego, focused on ML systems, the software and infrastructure that make machine learning models efficient and scalable. My recent work spans recursive self improvement for kernel development, high performance fused GPU kernels in CUDA and Triton, and scientific ML tooling built as DuckDB UDF extensions.
Before UC San Diego, I served six years in U.S. Army EOD and logistics, including leading a platoon that cleared more than 1,000 landmines and unexploded ordnance in South Korea. EOD gave me my first taste of deeply technical, multidimensional problems that reward ingenuity and contain tradeoffs. Software gives me a feel for that same challenge. After the Army, I moved into software engineering at Palo Alto Networks, where I led frontend integration of an AI copilot into Strata Cloud Manager. I came to UCSD to understand the systems beneath the software I was building.
Outside of that, I like running, paddle boarding and playing electric guitar. I'm currently into pop-punk, nu-metal and grunge.
PROJECTS
- Engineered an autonomous, Nsights guided evaluation harness for recursive self improvement, profiling and optimizing a k-means DSL kernel without human intervention.
- Achieved a 1.36× speedup on Qwen2.5-1.5B keys by dynamically bit-packing centroid IDs into the FP32 mantissa; this reduced register pressure, enabling block local accumulation that bypassed severe atomic_add HBM collisions (reducing atomic writes from per thread to per block).
- Analyzed top-k budgets for sparse attention; found they violated the top-p attention-mass for over 20% of heads motivating attention adaptive token budgets.
- Implemented zero copy FFI boundary in process memory sharing between C++ DuckDB User Defined Functions (UDF) and a Rust based Random Forest implementation via the Apache Arrow C Data Interface (ArrowArray / ArrowSchema).
- Implemented memory bounded in database TreeSHAP attribution streaming, dynamically deriving batch sizes to enforce a < 80 MB peak memory, computing an attribution double per sample, class and feature (197k sparse features) without OOMs, supporting browser WASM.
- Optimized sparse matrix scanning across 13M+ cells using DuckDB's UnifiedVectorFormat using std::string_view avoiding string materialization before interning as sparse COO representation. Used flat memory layouts and strides rather than Vec<Vec<Vec<f64>>> eliminating intermediate heap allocations and introducing cache line friendly access.
- Implemented and benchmarked a fused FP16 matrix multiplication, elementwise add, and ReLU kernel in Triton. Evaluated five tile configurations and achieved a 1.28× speedup over the PyTorch reference (0.265 ms vs. 0.339 ms) on an NVIDIA A10 for a 2048×2048×2048 workload.
- Implemented NCCL style distributed training primitives, binary-tree All-Reduce, point-to-point All to All, via Open MPI, to support data, tensor and expert parallelism (MoE).
- Took a BLIS style 5 loop setup to 97% of vendor BLAS (Arm Performance Libraries) GFLOP/s using cache blocked page aligned buffers, broadcast outer GEMM SVE intrinsics, and grid search autotuned params.
- Engineered a custom 4x16 microkernel utilizing 256 bit vector registers; 16 for accumulation, 8 for loading, and 8 for pipelining and predicates. Used non temporal stores to reduce cache pollution.
- Implemented cases in Loma's forward and reverse mode autodiff passes, extending existing intermediate representation (IR) mutators to transform program IR into first-order derivative compiled C functions.