I am an applied machine learning and AI engineer based in Dresden, Germany. I hold a German Diplom in Business Informatics (M.Sc. equivalent) from TU Dresden and have hands-on experience in industrial signal processing, production ML services, data-driven optimization, and local-first AI applications.
I work at the intersection of machine learning and software engineering—turning noisy data, domain constraints, and research prototypes into testable systems that people can use.
- 🏭 Industrial ML: time-series classification, anomaly detection, semantic segmentation, and HFCT signal analysis
- 🧠 Applied AI: cited RAG, structured LLM workflows, semantic knowledge graphs, and human-in-the-loop systems
- ⚙️ Optimization: multi-objective search, reinforcement learning, dynamic programming, and profit-aware decision pipelines
- 🛠️ Engineering: Python and TypeScript systems spanning desktop tools, APIs, web applications, plugins, tests, and CI
- Built a memory-efficient desktop workflow for engineers to inspect and label paired industrial sensor recordings containing up to 560 million samples without loading entire files into memory.
- Developed classification and segmentation workflows for partial-discharge signal analysis, including a deployed inference path for customer measurements.
- Built data-driven structural design software connecting predictive models, OpenSees analysis, and evolutionary optimization with interactive Pareto exploration.
- Led a three-person energy-storage optimization team to 350th place among 6,000+ entries (top 6%) using leakage-aware validation and profit-based model selection.
- Released Vault Coach, an open-source Obsidian learning assistant built around source-grounded answers, practice, mastery tracking, and reviewable AI suggestions.
An open-source, local-first learning assistant for Obsidian with cited RAG, semantic search, generated exams, inspectable knowledge graphs, mastery tracking, and review recommendations. AI-proposed relationships remain reviewable before they affect the learning map.
TypeScript Obsidian API RAG Local Embeddings Knowledge Graphs JSON Schema Vitest GitHub Actions
An end-to-end machine learning pipeline that turns electricity-price forecasts into valid battery charge/discharge schedules. The workflow combines time-aware validation, market and weather features, profit-based model selection, constrained strategy search, and reproducible submission checks.
Python pandas scikit-learn XGBoost Time-Series Validation Feature Engineering Decision Optimization
| Area | What I have worked on |
|---|---|
| Industrial ML & Signal Processing | Long-sequence HFCT visualization and labeling, partial-discharge classification, anomaly detection, semantic segmentation, model quantization, and production inference |
| Optimization & Applied Research | Offshore wind substructure design, OpenSees-based structural analysis, NSGA-II/GA search, Pareto clustering, and constrained bridge-layout optimization with RL and dynamic programming |
| AI & Knowledge Systems | Local-first RAG, source citations, semantic knowledge graphs, structured LLM outputs, user memory, human-review gates, and asynchronous AI workflows |
| Software Engineering | Python APIs, TypeScript applications, Obsidian plugins, PyQt desktop tools, Flask services, interactive visualizations, automated tests, and CI |
Machine Learning & Data
AI & Application Engineering
Engineering Tools
The language chart reflects public repository activity, not overall proficiency.
I am open to opportunities in Machine Learning Engineering, Applied AI, Data Science, Industrial AI, Optimization, and AI Software Engineering—especially roles where rigorous modeling and practical software delivery matter equally.
If you are working on industrial AI, intelligent engineering software, optimization, or trustworthy LLM applications, feel free to connect on LinkedIn or send me an email.