Applied AI for understanding and building real-world systems
I came to AI through physical systems like fluids and thermal dynamics, focused on understanding and modeling how things behave. Since then, each step has been about building the tools and perspective needed to apply AI more effectively to physical systems and scientific problems.
Along the way, I've also spent time working with founders and early teams — helping startups turn new ideas into working systems across domains like coding, mechanical engineering, healthcare, gpu kernel optimization, and voice AI. I enjoy being close to the point where ideas become products, and where technical choices matter early.
Applied AI across domains
I work at the intersection of applied AI, complex systems, and fast-moving product environments, with a focus on turning frontier models into reliable, production-ready tools.
Arup — engineering & sustainability
Thermal, energy, and microclimate systems, modeled with CAD, CFD, and machine learning, where models must be grounded in physical laws, explicit constraints, and verifiable assumptions.
Neural Concept — 3D geometric AI
Built the foundations for agentic and human-feedback-driven workflows, enabling aerospace, automotive and consumer products teams to explore and accelerate design using 3D geometric AI.
OpenAI — applied RL & evaluation
Led applied work on agent reinforcement fine-tuning and model evaluation, focused on reward modelling, graders, feedback, decomposition, and failure modes to improve agentic behavior over time.
Across roles, my work has centered on the same challenge: building learning systems that move fast without losing alignment with first principles, and helping teams — especially early-stage ones — make sound technical decisions as ideas scale into products.