University Machine Learning Competitions (Top-3 in Three Challenges)
I placed in the top three across three university Kaggle competitions spanning tabular machine learning, a PyTorch CNN trained from scratch, and few-shot LLM classification.
Result
Top-3 finish in all three university Kaggle challenges
I completed three University of Trier course competitions with reproducible single-file Python submissions.
For taxi-tip prediction, I engineered temporal and fare-derived features with HistGradientBoosting.
For image classification, I trained a custom VGG-style PyTorch CNN from scratch using data augmentation, AdamW, and cosine learning-rate scheduling.
For text classification, I built a few-shot pipeline using 4-bit Gemma-3-12B inference with similarity-based exemplar retrieval.
PolarsScikit-LearnPyTorchGemma-3-12B