Models, data pipelines, and scientific software for energy, nuclear safety, imaging, hydrology, and industrial process engineering.
Energy and physical systems.
Active projects across power grids, nuclear safety, volumetric imaging, and stakeholder-facing hydrology.
Distribution networks · Probabilistic ML
Probabilistic foundation models for distribution networks
Electricity distribution networks require models for a range of analysis and optimization tasks. A probabilistic foundation model can be fine-tuned for tasks such as optimal power flow and state estimation, providing predictions with uncertainty estimates.
My contribution
Developed the model and evaluation workflow, building on the GridFM approach initiated by IBM Research and adapting it to energy-research use cases.
Approach & validation
Learn a shared representation of distribution networks, then fine-tune it for individual tasks. Evaluate both prediction accuracy and uncertainty for each application.
Toolkit
PythonPyTorchFoundation modelsDistribution networksOptimal power flowState estimationProbabilistic modeling
Codes such as ASTEC can run slower than the accident they simulate, limiting their use for operator training and time-critical scenario assessment. A learned surrogate makes those progressions explorable in real time.
Simplified pressurized-water reactor represented by the ASTEC severe-accident simulator.
My contribution
Designed and evaluated the vessel-and-core surrogate with the KIT and ASSAS teams. The model outperformed a persistence baseline for held-out rollouts spanning 10.5 minutes of accident progression.
Approach & validation
A 27.2-million-parameter multimodal Transformer encodes vessel mesh cells, faces, and core elements separately, evolves them jointly in a shared latent space, and decodes them back into ASTEC-compatible outputs.
MRI captures non-invasive flow measurements, but turning them into a coherent, time-resolved three-dimensional flow field is a difficult inverse problem.
My contribution
Built the three-dimensional data and model pipeline with researchers at the TUHH Institute of Process Imaging.
Approach & validation
The workflow treats time-resolved volumetric reconstruction as an inverse problem and validates learned fields against experimental measurements.
An AI assistant helps users set up and run hydrology simulations, explore the outputs, and compare scenarios to inform water-management policy.
My contribution
Designed the interaction and tool-integration layer for the Solution Lab Rur–Erft digital-twin workflow.
Approach & validation
The assistant connects users to simulation and visualization tools, supporting scenario setup, execution and interpretation. Users guide the exploration and make the decisions.
Four years delivering data, simulation, and cloud tooling inside a DAX industrial company.
Computational Particle, Fluid and Reaction Modeling · BASF SE · 2021—2025
Plant-data, simulation, and cloud tools for production teams
Production teams need models for their troubleshooting, forecasting, and investment decisions.
My contribution
Analyzed plant data, modeled production assets using simulation software or data-driven approaches, supported troubleshooting and investment decisions, and built interfaces used to bring models into operational workflows.
Approach & delivery
Combined Python-based analysis and process simulation with containerized Dash and React applications deployed on Azure.
One focused body of work on validated particle-scale simulation, reduced-order modeling, and industrial process scale-up.
Doctoral research · Computational process engineering
Validated particle-scale models for industrial granulation
Industrial granulation runs for hours, while detailed particle simulations often cover only minutes. Bridging that gap makes particle-scale insight useful for process scale-up and product design.
My contribution
Designed and implemented numerical methods in C++ and MPI in OpenFOAM and LIGGGHTS/CFDEMcoupling, ran HPC studies, validated simulations against experiments and reference solutions.
Approach & validation
A reproducible validation workflow compared flow behavior, residence times, spray contact, and product-property predictions across apparatus scales and model variants.
