Machine Learning Systems Architect | AI Infrastructure Builder

Building the bridge between fundamental AI research and high-performance production systems. My core focus is on pushing the boundaries of hardware efficiency (TPU/GPU) for novel architectures, and industrializing software delivery.

• ChaosAI: Architecting a General Time-Series Foundation Model. Fusing Mamba-2 (SSM), JEPA, and Continuous Flow Matching (OT-CFM) to model chaotic dynamical systems.
  • Infra: Orchestrating 768 TPU v5p via Google TPU Research Cloud (TRC). Pure JAX, GSPMD sharding, and float32 accumulated bf16 kernels.

• AI / Compute: JAX, Flax, Optax, PyTorch, XLA Compiler Optimization, TPU Pod Topology.
• Backend / Systems: Python, C, Bash, IPC, Rust (Exploring).
• Frontend / Full-Stack: TypeScript, Angular, React Native, Tailwind.
• Infrastructure / Ops: GCP, Docker, CI/CD, FinOps, Data Lakes (Grain/ArrayRecord).

• "Optimize for insight per dollar, not for appearances."
• Hardware-aware design beats raw compute throwing.
• The execution is the only metric that matters.

📫 Contact: george-daniel.gherasim@ensta.fr | 🔗 ORCID: 0009-0009-3026-844X