Core expertise

Time-series forecasting, stochastic optimization, Bayesian modelling, scientific ML, uncertainty quantification.

Applied statistics / data science

• Time-series modelling (statistical and machine learning models).
• Bayesian and frequentist statistics (regression, hierarchical models, hypothesis testing).
• Extreme value analysis (return periods from 1 to 10,000 years, stationary and non-stationary processes, GEV, GP; emphasis on fitting to small datasets with Bayesian statistics).
• Deep learning: regression and image analysis (CNN).
• Risk assessment and decision theory.

Forecasting & experimentation in production

• Probabilistic/distributional forecasting: proper scoring rules (CRPS, Brier, pinball loss), quantile calibration (PIT, coverage), conformal prediction.
• Neural quantile regression: monotone quantile networks, distributional output heads (PyTorch).
• Time-series foundation models: zero-shot inference and LoRA fine-tuning (Chronos, TiRex, TimesFM).
• Risk-aware optimization: CVaR (Rockafellar-Uryasev) objectives, differentiable optimization layers, MILP/LP solvers (Gurobi, scipy, cvxpy).
• Experimental design: moving-block bootstrap, paired significance testing, Cohen's d effect sizes, Sobol sensitivity analysis, A/A tests, synthetic controlled-truth benchmarks.

Production & infrastructure

Docker, Kubernetes, MLflow, Kubeflow, CI/CD, Terraform, Databricks, Metaflow, Argo, Dagster.

Programming

• Python (15+ years): numpy, scipy, pandas, polars, scikit-learn, PyTorch, TensorFlow, LightGBM, cvxpy, PyMC, pyMC3, emcee, dynesty, catboost, sphinx, flask; functional and object-oriented, package development, testing, semantic versioning.
• Julia: Flux.jl, DifferentialEquations.jl, DiffEqFlux.jl, Turing.jl, NLOpt.jl, Zygote.jl.
• R: dplyr, rstan, evd, shiny, ggplot2.
• MATLAB: Statistics, Optimization and Neural Network toolboxes, UQLab.
• SQL, Git, Linux.

Data & systems

PostgreSQL, Snowflake, dbt, Feast, AWS (EC2, S3).

Numerical analysis / computational science

• Optimization: gradient-based (finite differencing, algorithmic differentiation, Newton-Raphson, BFGS, stochastic gradient descent) and gradient-free (Nelder-Mead, COBYLA, Bayesian optimization, multi-objective, heuristic methods).
• Finite element method for PDEs.
• Bayesian parameter estimation and model selection: MH-MCMC, variational Bayes (ADVI), nested sampling, NUTS, affine-invariant ensemble sampling.
• Scientific machine learning: combining ODE/PDE models with neural networks.
• Surrogate modelling of computationally expensive simulations.

Leadership

• Technical lead and proposal writer for multi-million-euro research and production projects.
• Led and managed research and engineering teams of 3-7 people.
• Mentored ML scientists, junior researchers, and 30+ MSc students.
• Ran internal workshops on software engineering, ML, and applied statistics.

Domain background (civil/structural engineering, PhD years)

FEM analysis (ANSYS, OpenSees, Diana, Atena, AxisVM, ConSteel, MIDAS Civil), GIS (QGIS), AutoCAD.