I work with industrial partners to turn methods developed for demanding scientific environments into practical AI systems. The focus is on domain-specific models that can run close to the instrument, work with limited or unusual data, and deliver results that can be validated in the physical world.

Domain-specific data Local deployment Measured performance

01

Precision reconstruction

Methods developed for particle detectors can transfer to tomography, industrial imaging, and other sparse sensor systems. The aim is fast, high-fidelity 3D reconstruction that turns raw measurements into useful information on site.
Reconstructing a complex particle-detector event from sparse measurements.

02

Edge perception

Large vision models can be distilled and optimized for robots, drones, and embedded systems. Running perception locally reduces latency, power use, and dependence on cloud infrastructure while keeping sensitive data under the partner’s control.
Real-time segmentation and detection for local deployment.

03

Generative materials discovery

Property-guided generative models can propose materials against partner-defined constraints. Coupled with laboratory validation, this creates a closed loop for exploring candidates for magnets, energy storage, catalysts, and other advanced materials.
A diffusion model generating candidate lithium–iron–oxygen crystal structures.