AI/ML Engineer building reliable intelligent systems from models to production.
I work across uncertainty-aware modelling, graph neural networks, reproducible experiments, and the engineering needed to make evidence usable.
- Reliable ML, uncertainty quantification, and scientific computing
- Graph neural network surrogates for transport simulation
- MLOps, evaluation pipelines, APIs, and containers
- Grounded RAG and local-first AI application engineering
- Python, PyTorch, scikit-learn, FastAPI, Docker, and pytest
| Project | Evidence boundary |
|---|---|
| Transport Policy Surrogate UQ | The thesis codebase, holding the submitted document as a frozen record alongside a working copy edited after submission, and saying which is which. Artifacts live in ml-surrogates-thesis-data. Supersedes ml-surrogates-thesis, now archived. |
| End-to-End MLOps Pipeline | Tested reference implementation with validated data, lineage, promotion gates, serving, and deterministic synthetic fallback. No production deployment is claimed. |
| InsureAssist RAG | Local engineering prototype with cited retrieval, FastAPI, Docker, and authored Kubernetes manifests. No completed cloud deployment is claimed. |
| Hydrology Uncertainty Quantification | Three-person TUM group coursework on calibration, sensitivity, and uncertainty propagation. Individual ownership of all results is not claimed. |
| Portfolio Platform | Next.js portfolio with typed factual governance, accessibility checks, security headers, and Playwright regression tests. |
My Master's thesis, Uncertainty Quantification for Machine Learning Models in Transportation Policy Analysis, was submitted at the Technical University of Munich on May 15, 2026. This is a submission statement, not a claim of degree conferral, defense, grade, or graduation.
Public identity, status, and contact facts were reviewed on August 20, 2026. The project rows were re-checked on September 2, 2026, when the thesis repositories were consolidated. Numerical claims remain bounded by the versioned artifacts and limitations in each repository.