mouad elalj

At OpenWorlds, I build simulation, validation, and monitoring systems for autonomous language-model agents operating on live markets. I created the backtest application, including historical ingestion, replay, config-driven runs, JSONL logging, portfolio simulation, and a live dry-run harness.

Previously at Hyperspace AI, I built distributed AI infrastructure for machine-learning workloads on contributed consumer hardware. I developed CUDA, Metal, and WGSL/WebGPU accelerator paths; made GPU kernels bit-exact on CPU and WebAssembly; and built Rust services for continuous batching, streaming inference, telemetry, and cross-platform GPU builds.

At Pyccel, I built its CUDA backend to translate annotated scientific Python into GPU kernels and memory management. I also ported numerical research code to GPUs, profiled slow paths, supervised interns, and taught programming labs.

I enjoy making heavy algorithms run fast and identically on hardware I do not control. My work spans Python, C, C++, Rust, CUDA, WebGPU, model serving, compiler tooling, and production services.