Research
Optimizing LLM-agent harnesses through causal execution graphs, cost-aware search, and transferable online policies for faster and more reliable agent execution.
- Causal execution graphs
- Constrained optimization
- Multi-fidelity search
- Online policy learning
Assignment and routing optimization across multimodal transport networks for resilient and cost-aware logistics decisions.
- Network optimization
- Assignment models
An automated and reproducible pipeline for constructing simulation-ready corridors between arbitrary origin-destination pairs, integrating OpenStreetMap and EPQS APIs to extract route alignments and horizontal and vertical terrain profiles from open-source geospatial data, with direct integration into ALTRIOS for energy and operational simulation.
- Workflow automation
- Automated data pipeline
- Geospatial modeling
- Signal smoothing
- Python
Developed a Fourier Neural Operator to learn spatial mappings from publicly available rail corridor features to official reference profiles, improving feature alignment, data consistency, and downstream simulation fidelity.
First-stage spatial optimization for the LIFTS project, developing a mixed-integer linear programming model for Parallel Systems' micro-terminal layout design under capacity and spatial constraints, with the objective of minimizing total system travel distance.
- Mixed-integer programming
- Terminal layout planning
- Heuristic algorithm
- Facility design
Discrete-event simulation and operational modeling of intermodal terminals to evaluate freight-handling performance under dynamic operating conditions. LIFTS has been integrated into ALTRIOS, an open-source digital-twin and simulation platform for modeling railway energy, safety, and operations. ALTRIOS has received two R&D 100 Awards, recognizing it among the year's most innovative technologies.
Reinforcement learning-based train dispatching to improve railway operational resilience under disturbances.
- Reinforcement learning
- Actor-critic algorithm
- Train dispatching
- Disruption management
Computer vision models for automated detection and segmentation of railway track defects from inspection imagery. The system was deployed on a railway inspection vehicle for field applications, and the associated patent (ZL202120098784.1) was sold to a railway technology company.
