Publications

Tests whether reinforcement learning can train a code model to generate a reusable solver for a family of constrained optimization problems.

A controlled study of structured multi-agent and two-agent workflows for requirements extraction, design decomposition, and executable physics-model generation.

An integrated interface for near-real-time performance exploration and automated CAD generation of gas-bearing turbocompressors.

A constrained all-at-once formulation for balancing efficiency, operating range, and robustness across an interdependent turbocompressor system.

A surrogate-assisted formulation that treats robustness as an explicit design objective and achieved a four-order-of-magnitude evaluation speed-up in the reported compressor study.