Transformer-NES
A mixture-of-experts transformer-based NES emulator constructed analytically from the 6502 ISA in C++26 and Python. Features an analytic weight compiler and a runtime with O(1) attention that drives a fully playable emulator.
Kyle Lukaszek
kəm-ˈpyü-təri like inverse problems & ml
Profile
Hi, I'm Kyle. I like to work at the boundary of machine learning, mathematical modelling, and graphics. My research interests span perceptual modelling, representation learning, and programming GPUs to simulate light transport. In my free time, I like solving puzzles, reading technical books, and playing games.
Skills
- Languages
- C
C++ Rust Python Julia Haskell Slang - ML / Scientific
- PyTorch
JAX MLX SciPy - GPU / Graphics
- CUDA
Metal WebGPU Vulkan DX12 - Tooling
- Nsight
gdb/lldb valgrind Slurm git/jj uv
Achievements
- ML Puzzle — "Puzzle" · Jane Street · 1 of 38 final solvers
- ML Puzzle — "Dropped a Neural Net" · Jane Street · 1 of 100+ solvers
Projects
Kiln
Render hardware interface for Vulkan 1.4 and Metal 4 in Rust. Supports acceleration structures and compute, mesh, vertex, and pixel shader pipelines. Powers Spectra, my compute-based spectral path tracer for USD scenes.
SDL3 Haskell
Lead maintainer for the Haskell bindings of SDL3, the native application development library. Comes with 50+ examples, including 30+ cross-platform GPU rendering examples spanning Vulkan, DX12, and Metal.
Neumours
Implicit neural tumour representations in JAX and Slang. On BraTS 2023+, the model reaches 0.823 Whole Tumour Dice — 92% of nnU-Net — with 118× fewer parameters and 20× faster inference.
Includes a differentiable Slang raymarcher for visualising segmentations.
WebGPU Projects
Real-time LiDAR rendering at 100M+ points on SensatUrban, GPU-accelerated colour-space visualizations, interactive scene-graph applications, and SDF geometry renderers running entirely in the browser.
Experience
Teaching Assistant — Parallel Programming, Operating Systems, Software Systems
- Parallel Programming (CIS*3090): Returning for F26 after teaching Pthreads, OpenCL, MPI, and CUDA during F25 office hours.
- Operating Systems (CIS*3110): Taught weekly labs on Unix, bash, scheduling, concurrency, and paging.
- Software Systems Development (CIS*2750): Supported instruction in C systems programming, CPython FFI, and SQL.
Research Assistant — HCI: Hardware & Statistical Modelling
Developed a cross-platform C driver for a JETI Spectraval 1501 spectrometer, replacing a Windows-only FTDI workflow and reducing experiment iteration time by over 50% while retaining full instrument accuracy. Co-authored peer-reviewed work accepted to Graphics Interface 2024.
Research Assistant — NLP: Pre-training & Post-training
Built statistical and NLP pipelines for a large Twitter corpus, engineering psycholinguistic, sentiment, personality, and behavioural features across thousands of users. Trained and fine-tuned LSTMs and Transformers with CUDA/PyTorch; performed time-series and causal analyses using ARIMA, stationarity testing, and statistical effect estimation.
Publications & Demos
Seeing Your Screen Outside: Surround Effects on Brightness Discrimination
Modelling The Effects of Bright Environments on Colour Perception
Portable GPU-Accelerated Colour-Space Visualization
Education
Master of Science in Computer Science
Perceptual modelling, machine learning, generative models.Mathematical Modelling · Numerical Linear Algebra / Scientific Computing · AI · HCI · Research Methods
Honours Bachelor of Computing
Advanced Calculus · Differential Equations II · Graph Theory · Linear Algebra · Statistics · Algorithms · Compilers · Computational Intelligence · Graphics · Operating Systems · Parallel Programming