Open source
Every claim below has an implementation you can clone and a measurement you can re-run. If a result is not reproducible from this code, it does not belong here.
Complete neural CPU executing ARM64 on GPU
A complete neural CPU where every ALU operation is a trained neural network achieving 100% accuracy on 32-bit integer arithmetic. Includes a full neural operating system (neurOS), GPU-native Metal compute kernel with 130+ ARM64 instructions, a 25-command multi-process UNIX shell, and structural timing side-channel immunity.
24 trained models (13 ALU + 11 neurOS) with exhaustive verification of all sub-component inputs
Neural multiplication (O(1) LUT) is faster than neural addition (O(log n) CLA)
25-command shell with fork/pipe/wait/dup2, compiled C on Metal GPU
AES-128 on GPU with zero timing variance (σ=0.0 across all runs)
465K lines of Rust - 14 quantum domains - native Metal GPU acceleration
A full quantum computing platform with a Rust core and native Metal GPU acceleration. 14 specialized domains covering error correction (37 modules), tensor networks, quantum ML, drug discovery, quantum trading, and more. One of only ~3 quantum tools in the world with Apple GPU support - and the only one with a complete research toolkit built on top.
Raw Metal compute shaders with UMA zero-copy dispatch, thermal-aware scheduling, and auto-backend selection
Error correction, tensor networks, noise models, algorithms, quantum ML, chemistry, backends, circuits, networking, physics, applications, measurement
Molecular fingerprints (ECFP4/6, MACCS), ADMET prediction, Lipinski filtering, QED scoring
Volatility surfaces, regime detection, quantum walk momentum, entropy-modulated Kelly criterion
Found a bug? Have an improvement? We welcome contributions from the community. All projects are released under permissive licenses to encourage collaboration.
VIEW ALL REPOSWant to understand the research behind these projects? Explore our research