At InstaDeep, I re-architected the InstaGeo PyTorch pipeline in JAX/Flax to deliver a 300× inference speedup and stronger multi-GPU scaling. I also build scalable BioAI deployment workflows, reinforcement learning agents, power-system models, and autonomous QA agents.
I've worked across JAX, distributed GPU/TPU training, and scientific computing, including authoring Dualpy, a pure-Python automatic differentiation library. I also teach graduate reinforcement learning, covering PPO and TRPO.
