Resume | Matteo Gätzner – Blog & Portfolio

Education

ETH Zürich (Zürich, Switzerland) | 09/2023 – 02/2026

GPA: 5.4/6.0
Relevant Courses: Conformal and Probabilistic Prediction, Probabilistic AI, Reliable and Trustworthy AI.
Master’s Thesis: Anytime-valid Neural Uncertainty Quantification for SPECT Imaging
Thesis Details: Designed and implemented Transformer-U-Net, diffusion models, and confidence sequences in Python using PyTorch, Hugging Face, and JAX, with W&B for experiment tracking and hyperparameter optimization, to deliver reliable uncertainty estimates (Best Grade).

Technische Universität Berlin (Berlin, Germany) | 10/2019 – 03/2023

GPA: 1.4 / 5.0
Bachelor Thesis: Using the GDReG for Maximum Likelihood Training of Energy-Based Models
Thesis Details: Accelerated training of normalizing flows, achieving the same performance with 5x fewer epochs using PyTorch, pandas, NumPy and Matplotlib (Best Grade).

Wirtschaftsgymnasium Offenburg (Offenburg, Germany) | 09/2011 – 06/2017

Roche (Basel, Switzerland) | 03/2024 – 09/2024

Automated an enzyme phenotyping workflow across pipetting robots and LC-MS, cutting manual setup time by 4 hours per run and reducing overall analysis turnaround from 6 months (external) to under 2 weeks in-house.
Built and optimized drug-drug interaction predictors with GNNs, CatBoost, and other ML models, achieving an MCC of 0.8 through cross-validation and hyperparameter tuning using Python, PyTorch, scikit-learn, and W&B.

Fraunhofer Heinrich-Hertz Institute (Berlin, Germany) | 05/2022 – 10/2023

Developed and deployed a deep-learning-based heat-island simulator on a Linux server for Dresden and Langenfeld, delivering city-scale predictions at 10 m resolution with an MAE of 0.6°C (Website: www.klips-projekt.de).
Developed a heat and humidity map forecasting model using a custom attention-based geospatial interpolation module, leveraging PostgreSQL for data management to achieve an MAE of 0.8°C for temperature and 8% for relative humidity.
Engineered a Docker-containerized FastAPI REST service with Nginx to serve model outputs, delivering 49 large-format (2300x1700 px) forecast heatmaps with full bundle transfers in under 2 seconds.

Technische Universität Berlin (Berlin, Germany) | 10/2021 – 12/2022

Prepared materials and graded assignments for core graduate and undergraduate ML courses for 50-100 students.
Developed a coding-based exam, assessing student proficiency across Python's data science stack (NumPy, pandas, scikit-learn).

TechLabs Berlin (Berlin, Germany) | 09/2020 – 06/2022

Provisioned and managed software development infrastructure, including GitHub repositories, for hundreds of students across four tracks (UX Design, Data Science, Web Dev, AI), streamlining project setup and improving contributor velocity.
Enhanced and maintained a JavaScript-based student management application, automating user provisioning and resolving technical support requests for the entire student cohort.

TPR English School (Zhuhai, China) | 06/2018 – 06/2019

Planned and delivered English lessons to beginner and intermediate students.
Designed lesson plans, exercises, and assessments focusing on reading, writing, listening, and speaking skills.

Patti’s Hire & Absolute Party Hire (Sydney, Australia) | 06/2017 – 04/2018

Delivered various types of furniture and set up marquees for events and parties (weddings, corporate events, festivals).
Coordinated logistics and scheduling to meet tight event timelines.

Author (Zürich, Switzerland) | 09/2025 – 01/2026

Developed a principled framework for uncertainty quantification in CT reconstruction using sequential likelihood mixing, establishing theoretical coverage guarantees for deep learning models (U-Net, Diffusion) under Poisson noise.
Utilized the framework to detect hallucinations, geometric inconsistencies and visualize interpretable uncertainty maps.
Submitted to the International Conference on Machine Learning (ICML) 2026. Preprint: arxiv.org/abs/2602.05812.

Author & Maintainer (Zürich, Switzerland) | 01/2025 – Present

Built high-performance MCMC samplers in Rust, implementing a suite of algorithms (HMC, NUTS, Gibbs) and diagnostics.
Published on GitHub with an extensive test suite, CI/CD via GitHub Actions, and comprehensive documentation.

Author & Maintainer (Zürich, Switzerland) | 2026 – Present

Built an open-source library that automatically generates high-quality Anki flashcards from EPUB documents using LLMs, providing a highly cost-effective study tool.

Operating Systems: Linux, macOS, Windows.
Programming Languages: Python, Rust, C, SQL.
Developer Tools: Git, CMake, Make, Docker.
Frameworks & Libraries: PyTorch, Hugging Face, JAX, WandB, MLflow, FastAPI, Geopandas, Rasterio.
Databases & Web Technologies: PostgreSQL, Nginx.
Other Skills: Linux CLI tools, CI/CD practices, REST API design.