I'm software consultant at Ab Initio software with a background in statistical physics, materials simulation, and deep learning. My PhD and postdoctoral research applied neural networks to inverse design problems in 2D materials and soft robotics — finding patterns where classical methods couldn't scale. My recent personal project is building practical ML systems: pipelines, multi-agent frameworks, and production-grade models.
Autonomous multi-agent system for accelerated discovery of optimal kirigami structures. Agents search a 32,768-structure design space using Gaussian Process + Kriging Believer or NN + ε-greedy active learning — finding targets in 8 cycles vs 147 for random search.
Local multi-agent orchestration framework with sandboxed tool execution. Agents collaborate on tasks using locally-hosted LLMs — no cloud dependency.
End-to-end ML pipeline for ad targeting: feature engineering, model training, evaluation, and serving — structured for production deployment.
Deep learning for inverse design of 2D kirigami metamaterials. Neural networks predict cut patterns that achieve target mechanical properties — replacing expensive FEM sweeps.