M.Sc. Data Science @ TU Dortmund · B.Sc. Computer Science (GPA 18.93/20, ≈1.3 German scale) · Bochum, Germany
I build machine-learning, data-quality, and probabilistic-modelling projects in Python, with 3+ years as a university teaching assistant in mathematics and numerical methods. Open to Werkstudent and internship roles in Data Science & ML in Germany.
- Used-Car Price Data-Quality Pipeline — containerized ETL with two-layer data validation (pandera + business rules), PSI drift monitoring, SQL storage and a Streamlit dashboard; on 426k real Craigslist listings it quarantined 43% of rows and trained a Random Forest to MAE 5,833 / R² 0.53.
- Adaptive Backtracking vs. Adam — controlled 64-run sweep (batch size × label noise × optimizer) under a fixed compute budget; finding: batch size, not label noise, is the binding constraint — adaptive learning-rate backtracking is competitive full-batch and collapses under mini-batching, explained by revert counts and learning-rate pinning.
- Protein Secondary Structure via Neural Posterior Estimation — M.Sc. team project; I built the data & ground-truth layer. Two-state HMM simulator + exact Forward–Backward posteriors (
hmmlearn) generating per-residue training targets for a Bayesian neural posterior estimator, plus the PDB 1A7F human-insulin validation set (Biopython) the model was benchmarked against (r = 0.855, chain B). 45 unit tests. - Image Recognition: SIFT vs. CNN — benchmarked CNN (VGG16) vs. SIFT features across MNIST / CIFAR-10 / Fashion-MNIST (~77% vs. ~47% accuracy). (B.Sc. thesis)
- Credit-Card Fraud Detection — K-means + Random Forest; 0.93 precision / 0.80 recall on the fraud class (~0.2% positive).
- COVID-19 Research Topic Modeling — LDA & NMF + K-means on a ~10,000-article corpus.
- Rainfall Forecasting — SMOTE + SVM on imbalanced weather data; ~80% accuracy, 0.77 recall on rain days.
- FIFA Player Analysis — PCA + K-means playstyle clustering.
- Languages: Python, SQL (PostgreSQL / SQLite), C++, C
- ML & Data: scikit-learn, NumPy, Pandas, TensorFlow / Keras, OpenCV — classification, clustering, PCA, NLP (LDA/NMF), CNNs, SMOTE
- Probabilistic modelling: Hidden Markov Models, exact inference (Forward–Backward), hmmlearn, simulation-based inference, Monte-Carlo validation
- Data engineering & quality: pandera validation, drift monitoring (PSI), ETL design, SQLAlchemy, Docker / docker-compose, Streamlit, pytest + GitHub Actions CI
- Foundations: numerical linear algebra, optimization, statistics, algorithms