Now (May 2026): Junior Data & ML Engineer at Epiq Energy ApS, Copenhagen — applied ML in the European energy domain.
Machine learning engineer with a mathematical and statistical foundation, working across deep learning, time-series forecasting, high-performance computing, and scientific data analysis. B.Sc. in Machine Learning & Data Science from the University of Copenhagen (DIKU), with coursework in measure theory, probability theory, and high-performance programming. I build systems that bridge theoretical rigour and production-grade engineering — from CUDA-accelerated neural networks in C to walk-forward backtesting pipelines for European electricity markets.
| Mathematics & Statistics | Probability Theory & Statistics, Linear Algebra, Lebesgue Integral & Measure Theory, Models for Complex Systems |
| Machine Learning & AI | Introduction to Machine Learning, Advanced Deep Learning, Medical Image Analysis |
| Computer Science | Algorithms & Data Structures, High-Performance Programming (C, CUDA, OpenMP) |
| Applied Research | Satellite Image Segmentation (collab. with IGN), BSc Thesis with DMI |
May 2026 – present
Applied ML in the European energy domain.
January 2024 – January 2026
- Built Python analytics tooling and SQL data models for operational insights across IT/OT environments in a regulated pharmaceutical setting.
- Performed statistical analysis on incident and root-cause trends to support reliability improvements.
- Developed an LSTM-based recurrent neural network for predictive maintenance, integrated with OPC-UA data streams.
- Deployed monitoring solutions (SCOM, SquaredUp, Grafana) for real-time IT/OT infrastructure observability.
- Worked across local IT infrastructure, automation (OT) systems, and MES in a GMP-regulated pharmaceutical manufacturing environment.
European Electricity Price Forecasting — eu-electricity-forecasting
XGBoost-based day-ahead price forecasting pipeline for 21 European bidding zones with walk-forward backtesting and trading strategy evaluation.
- Dual forecasting modes: single-model and per-hour XGBoost with lag features respecting market gate closures.
- Feature engineering from heterogeneous sources: Energinet, ENTSO-E Transparency Platform, and Open-Meteo weather data (temperature, wind speed, solar irradiance).
- Walk-forward backtesting engine with transaction cost modelling across multiple bidding zones.
- 108 unit tests, data completeness gates, imputation audit logging, and model drift detection.
Multi-Language ML Benchmark — ml-language-playground
Systematic benchmarking of neural networks (MLP, CNN/LeNet-5) and a broader algorithm catalogue (regression, tree ensembles, sequence models, Gaussian Process, HMM, unsupervised) across C, Rust, and Python — CPU and GPU backends.
- Custom CUDA kernels, cuBLAS FFI bindings (Rust), cuDNN integration, and OpenMP parallelisation.
- Peak GPU throughput: 8.92M samples/s (C CUDA), 8.21M samples/s (Rust cuBLAS); 5–15× GPU advantage over CPU at scale.
- Mathematical documentation of forward/backward propagation, im2col convolution, and pooling operations.
- Automated benchmarking pipeline with reproducible caching and successive-halving hyperparameter tuning.
Satellite Image Segmentation for Arctic Mapping — Automatic-Satelitte-Island-Discovery
Project outside course scope at DIKU, in collaboration with the Department of Geosciences and Natural Resource Management (IGN) at the University of Copenhagen — semantic segmentation of land regions along Greenland's coastline.
- U-Net architecture (PyTorch, segmentation_models_pytorch) on high-resolution 4-band PlanetScope imagery; test-set accuracy 0.92, IoU 0.53.
- Geospatial preprocessing with Rasterio and GeoPandas; shapefile-based mask generation from IGN-provided ground truth.
- Trained on an NVIDIA RTX A6000 GPU cluster (CUDA 11.8); optimised inference for large-scale geospatial datasets.
Automated Quality Control of Climate Data (BSc Thesis) — Climate-Data-QC
Bachelor's thesis in collaboration with the Danish Meteorological Institute (DMI).
- Designed and benchmarked three structurally distinct ML architectures for faulty sensor detection.
- Analysed DMI's ETL pipelines and manual validation workflows to identify automation opportunities.
- Reduced manual overhead for climatologists through automated anomaly detection on Greenlandic weather station data.
Hidden Markov Model for Visual Attention Analysis
Probabilistic graphical model for simulating and analysing visual attention patterns from neural spike data (course project: Models for Complex Systems).
- Forward simulation, exact inference (variable elimination, message passing), and approximate inference (logistic regression).
- Hard-assignment EM algorithm for parameter learning; validated on simulated and real-world neural datasets.
Native X11 MCP Server — x11-mcp
Bare-metal Linux desktop automation server for Claude Code, using python-xlib, python-uinput, and AT-SPI.
- Framebuffer capture, kernel-level input simulation, and accessibility-tree integration.
- Extended with voice control pipeline (
x11-mcp-voice): openwakeword → NVIDIA Parakeet STT → Claude → piper-tts.
| Languages | Python, C, SQL, R |
| ML / DL | PyTorch, PyTorch Lightning, TensorFlow, Scikit-Learn, XGBoost, OpenCV |
| Methods | Deep learning (CNNs, LSTMs, U-Net), probabilistic graphical models (HMMs, EM), ensemble methods, time-series forecasting, anomaly detection |
| HPC / Systems | CUDA, cuBLAS, cuDNN, OpenMP |
| Data Engineering | ETL pipelines, Pandas, NumPy, Rasterio, GeoPandas, Alteryx |
| Visualisation | Matplotlib, Plotly |
| Infrastructure | Linux/UNIX, Git, Docker, MLOps |
- Teaching Assistant — Empirical Methodologies & Theory of Science (2021–2023). Research methodology and academic reasoning.
- Teaching Assistant — Introduction to Programming (2021). Python fundamentals for first-year students.
- Student Mediator — ML & Data Science programme representative at DIKU. Curriculum feedback and student outreach.
- Email: j.brons.christensen@gmail.com
- LinkedIn: linkedin.com/in/johannes-broens-christensen