blurb

Hi! I’m an ETH Statistics graduate looking for a hands-on ML or data science role in early-stage startups. I specialize in building scalable, interpretable AI systems always with a focus on real-world impact. Recent work ranges from tabular foundation models and sparse autoencoders to explainable deep learning. I conduct cutting-edge research (e.g., NeurIPS 2025 publication on in-context counterfactual reasoning) and find solutions to applied problems (like predicting NCAA March Madness outcomes with TabPFN). Below are two recent projects that showcase my approach.

1. Deep Learning-Based Pneumonia Detection and ECG Time Series Classification

• Vision Models: CNN + Grad-CAM/Integrated Gradients and Vision Transformer for pneumonia detection
• Time Series Models: RNN and Transformers for ECG time series classification and representation learning
• Task: Validated vision models with saliency maps; compared time series prediction performance
• Emphasis: Focused on explainability (Grad-CAM, Integrated Gradients, UMAP) and transfer learning
• Code: Check out our repo at GitHub

2. Sparse Autoencoders for Interpretable Features

• Paper: Explored feature intervenability in language models
• Goal: Orthogonal, intervenable features for improved identifiability and interpretability
• Innovation: Low-rank adapted language models to sparse autoencoders under orthogonality regularization
• Tools: PyTorch, HuggingFace, custom sparse autoencoder architectures
• Code: Find our code and more details on model adaptation on GitHub

Technical Skills Python (PyTorch, HuggingFace, Pandas, vLLM, TorchSDE, Transformers) and R (lme4, glmnet, rshiny)

Résumé Click here for my current CV