I am a Senior Machine Learning Software Engineer.
Focused on reinforcement learning, AI infrastructure, and building reliable and scalable software for AI systems.
- gym-puddle: Off-policy PAC algorithm implemented on the Puddle World Gymnasium environment using TorchRL
- proprio: Unsupervised, uncertainty-aware perception for a 7-DOF robot arm; classifies each lidar reading as self, background, or anomaly, without any geometry or kinematics.
- AlphaEx: Sweep parameters and dispatch thousands of Slurm jobs from one Python script.
- internals: Interactive, first-principles tutorials for modern AI systems & system components.
- Speculative Decoding: Interactive walkthrough of how LLMs emit several tokens per forward pass.
- nabla: Educational numpy implementations of 15 optimizers (SGD → Muon), animated on a 2D saddle & benchmarked on matrix LS.
- priori: Interactive marimo benchmark of TabPFN v2 — a tabular foundation model that predicts in-context, with no training — against tuned XGBoost & AutoGluon on churn and credit tables.
- minitorch: Minimalistic deep learning framework rebuilt from scratch: autodiff, tensors & a neural-net stack across NumPy, Numba-parallel CPU, & CUDA backends.
- kairos: Reinforcement learning when the environment won't wait: a bounded-compute PPO agent acts on a stale policy while each update computes, dropping the experience it's too busy to process — testing whether extra compute is better spent on more epochs or more fresh data.