Summary
Senior engineer working on CUDA math libraries, compiler integration, and agent-driven kernel development, with research in numerical computing and scientific machine learning.
Technical Focus
CUDA, C++, Python, GPU performance engineering, numerical software, compiler and IR development, JIT/LTO, MLIR, LLVM, SIMT programming, NEON, SVE, scientific computing
Experience
Senior CUDA Math Libraries Engineer | NVIDIA
- Developed a framework for AI-agent-driven CUDA kernel generation, profiling, optimization, modification, and autotuning.
- Developed modular, agent-friendly domain-specific languages for algebraic code transformations and code reuse.
- Developed CUDA Tile integration and generalized autotuning with SIMT interoperability, cuTile Python, NVCC, LTO IR, and compiler technologies.
- Built developer productivity tooling including harnesses, CLIs, reusable setups, sandboxing, onboarding guides, and GPU CI monitoring and triage.
- Studied GPU I/O topologies, algorithmic changes, and simulated performance to inform hardware/software co-design.
CUDA Math Library Engineer | NVIDIA
- Optimized CUDA math-library and cuFFT performance through memory-bandwidth work, asynchronous I/O strategies, and code generation.
- Used JIT LTO to generate optimized kernels at runtime for particular user problems while reducing required precompiled kernels.
- Optimized NVPL CPU math libraries on Grace CPUs with NEON and SVE vectorization.
Research Assistant (HPC) | Innovative Computing Laboratory, University of Tennessee, Knoxville
- Developed GPU compute kernels for MAGMA and SLATE for scientific and machine-learning applications on supercomputers.
- Optimized MAGMA BLAS for AMD GPUs; related work appears in a coauthored IEEE HPEC 2020 paper.
- Designed SABATH, a platform for reproducible scientific machine-learning applications following FAIR principles; related work appears in a coauthored IEEE HPEC 2022 paper.
Software Architect | InterGauge
- Provided technical leadership for an education-technology startup focused on interactive educational AI assistants and classroom focus metrics.
Machine Learning Intern | NVIDIA
- Researched hardware acceleration for neural-network training and inference through DSLs, intermediate representations, kernel generation, and polyhedral GPU compilation.
- Trained and evaluated models for code generation and optimization-pass selection.
Additional Experience
Research Assistant (HCI) | University of Tennessee, Knoxville
- Developed Avocat, a terminal tool that interprets error output and suggests shell-command solutions using generative text models and semantic vector search.
- Optimized graph-database queries for analytics over a terabyte-scale source-code dataset.
Machine Learning Consultant | Qardian Labs
- Trained and validated a model for heart-disease identification and anomaly detection, with a web interface for risk assessment and further-investigation decisions.
Research Intern (HPC) | Oak Ridge National Laboratory
- Contributed with Thomas Hill to SimpleSummit, a portable NVIDIA Jetson cluster used to explain distributed computing near Summit at ORNL.
- Developed FractalExplorer, a distributed GPU-accelerated interactive visualization for Mandelbrot and other fractals on SimpleSummit.
- Implemented and ran benchmarks to measure Lustre parallel-filesystem scalability on the Titan supercomputer.
- Assisted with fabrication of the SimpleSummit cluster’s 3D-printed enclosure for eight NVIDIA Jetson boards.
Software Consultant | Agilaire LLC
- Designed and implemented pilog, a hardware/software air-quality data logger with sensor collection, time-series storage, visualization, and programmatic access.
Publications
Piotr Luszczek and Cade Brown. Surrogate ML/AI Model Benchmarking for FAIR Principles’ Conformance. IEEE HPEC, 2022.
Cade Brown, Ahmad Abdelfattah, Stanimire Tomov, and Jack Dongarra. Design, Optimization, and Benchmarking of Dense Linear Algebra Algorithms on AMD GPUs. IEEE HPEC, 2020.
Talks & Presentations
SABATH: ML Surrogate Platform
- Presented SABATH’s approach to reproducibility in scientific machine-learning models.
Challenge #6: Sentinel-3 Satellite Dataset
- Introduced the SMCEFR dataset and data challenge for participating researchers.
MAGMA to hipMAGMA: Adding HIP Compatibility to a Scientific C++ Library
- Demonstrated HIP portability for CUDA scientific libraries targeting AMD GPUs.
Research Service
Dataset Sponsor | Smoky Mountain Conference, ORNL | 2022
Research
Cade Brown. OEIS A267263, 2016.
SMCEFR satellite-data contribution with Piotr Luszczek: a 1024 × 1024 RGB subset derived from Sentinel-3 OLCI EFR data; Cade Brown is identified as corresponding author.
Selected Projects
mycc: public C-subset compiler and reimplementation using Bison, Flex, and LLVM, with an optional ORC JIT path.
kscript: Dynamic language implemented in C, with arrays, linear algebra, FFTs, and a WASM REPL.
CARVE: a collaborative RISC-V emulator and visualizer for the web, developed with Gregory Croisdale under Stephen Marz.
Blok: Built a C/C++ voxel game with world generation, chunk and mesh rendering, and input handling using OpenGL, GLFW, AssImp, FreeType, and PortAudio (2020).
Personal website and research blog: Technical writing, tutorials, and digital artwork, maintained since 2015.
Open Source Contributions
MPFR | Open Source Contributor | 2016–2017
Leadership
FIRST Robotics #3966 | Lead Programmer | 2015–2018
- Designed and implemented software for semi-autonomous robots competing in FIRST Robotics events.
Education
Bachelor's degree, Computer Science | University of Tennessee, Knoxville
- FIRST Robotics Alumni Scholarship