Apple Neural Engine (ANE) LLM inference engine — reverse-engineered private APIs, Metal GPU shaders, hybrid ANE+GPU+CPU on Apple Silicon. 32 tok/s matching llama.cpp, 3.6 TFLOPS fused ANE mega-kernels.

A high-performance C++ simulation framework for modeling biolocomotion and premelting dynamics in ice using finite difference methods. This implementation numerically solves coupled partial differential equations for chemotaxis and diffusion: via OpenMP SIMD and via Metal for Apple Silicon.

High-performance C++ simulation framework for modeling active Brownian particles under cylindrical confinement in 3D space. Features dual implementations with standard OpenMP parallelization for cross-platform compatibility and Metal GPU acceleration for Apple Silicon, achieving up to 27× speedup for large-scale systems (N≥1000 particles).

A Ruby C extension providing native macOS window management and Metal GPU acceleration for graphics applications.

First native Apple Silicon (MLX) port of Diffusion Policy (RSS 2023 Best Paper). 6 policy variants, 472 tests, Metal GPU verified. Train and run visuomotor diffusion policies on M-series — no CUDA required.

Production RAG system for scientific literature synthesis with SPECTER2 embeddings, Metal GPU acceleration, multi-LLM support, and automatic BibTeX citations.

Comprehensive VAE performance benchmark comparing PyTorch vs TensorFlow on Apple Silicon (M1/M2/M3). Quantifies training speed, memory efficiency, and Metal GPU utilization across Python versions to guide framework selection for ML prototyping and production deployment.

LeRobot-MLX: HuggingFace LeRobot ported to Apple MLX for native Apple Silicon robotics policy training & inference. 10 policies, 739+ tests, Metal GPU accelerated.

GPU-accelerated robot motion planning on Apple Silicon. Port of NVIDIA cuRobo (CUDA) to MLX — real-time collision-free trajectory generation on M-series Macs.