Technical diagnosis
Review the decision, assumptions, existing data, code, and observed failure modes.
DeliverableA written risk map and recommended evidence plan.
Verified agentic discovery and design
Improve thermofluid systems—with quantified uncertainty.
I help industrial R&D teams improve heat-transfer and multiphase-flow systems. Agentic scientific workflows combine open-source CFD, experiments, and Bayesian optimization; verification and uncertainty quantification show which improvements can be trusted.
Best-fit assignments
When a design decision needs more than another simulation.
- Topology, shape, or parameter optimization where each CFD evaluation is expensive.
- Conjugate heat-transfer and thermal-management systems with geometric or operating constraints.
- Multiphase filling, wetting, mixing, and transport problems with uncertain or incomplete models.
- CFD workflows that need verification, reproducibility, or transfer into an engineering team.
Engagement model
A staged engagement, with evidence at every gate.
The first task is to define the decision and determine which evidence is actually needed—not to commit prematurely to a particular model or tool.
Feasibility and baseline
Reproduce or establish a verified baseline and test whether the modelling route can answer the question.
DeliverableAn auditable go/no-go result with explicit validity limits.
Build and transfer
Develop the method or workflow in a tested repository and transfer it to the client team.
DeliverablesSoftware, V&V evidence, reproducible cases, documentation, and technical training as required.
Every project begins with the engineering decision and its acceptance evidence. Reusable evaluators, tests, evidence records, and orchestration components are standardized when the application boundaries support responsible reuse.
Technical scope
Focused work around the decision that needs stronger evidence.
Assignments can be diagnostic, collaborative, or delivery-oriented, depending on the maturity of the existing model and team.
01
Model development
Translate a thermofluid question into governing assumptions, measurable quantities, and a defensible simulation strategy.
02
Verification and validation
Expose numerical error, force imbalance, mesh sensitivity, and validity limits before they become product risk.
03
CFD + ML integration
Connect solvers, data systems, surrogate models, and heterogeneous compute without brittle file-based handoffs.
04
Bayesian optimization
Use uncertainty-aware search when simulations and experiments are too costly for brute-force exploration.
05
Research software, data & provenance
Connect numerical methods, data lineage, automated tests, reproducible workflows, and documentation so a team can inspect, extend, and validate every result.
06
Agentic scientific workflows
Use client-approved language models and agent systems to orchestrate numerical-method development, simulation, testing, documentation, and audit; accept results only through explicit V&V, provenance, reproducibility, and human-review gates.
Evidence today
Scientific depth with explicit boundaries.
- OpenFOAM work since 2007 and co-authorship of The OpenFOAM Technology Primer.
- Five industry-funded PhD projects at TU Darmstadt across Mathematics and Mechanical Engineering—four with Bosch Corporate Research and one with Rimac Automobili.
- A four-study Bosch microfluidics evidence chain.
- Peer-reviewed numerical methods and maintained open software.
- Published CFD–machine-learning infrastructure and reproducible workflows.
Project governance: each engagement defines its operating envelope, contracting route, inputs, deliverables, acceptance criteria, schedule, background IP, ownership or licensing, data boundary, publication and reuse rights, and handover before implementation begins.
Commercial and data boundaries
Agree the operating rules before the workflow touches client material.
Technical ambition is useful only when confidentiality, ownership, tool access, and acceptance authority are explicit.
- Contracting
- The contracting party, scope, delivery capacity, and separation from university activity are confirmed before work begins.
- Client data and code
- Repository access, processing location, retention, confidentiality, and handover are agreed in writing.
- AI tooling
- Models, data routing, and tool permissions are disclosed and agreed. Confidential client material is not sent to an external AI service without explicit agreement.
- Engineering authority
- Model assumptions, acceptance criteria, and release decisions remain under human engineering review.
From first email to a defined project
Start with a design brief.
Describe the system, objective, constraints, operating envelope, available evidence, and desired timeframe at a level you are comfortable sharing. If technical details require protection, we can establish an NDA before the substantive exchange.
- Initial screenI check technical fit and whether a useful first step can be defined.
- Scoping conversationWe identify the smallest evidence package that could change the decision.
- Written project scopeContracting, confidentiality, IP, deliverables, acceptance criteria, schedule, and handover are made explicit before work begins.