Chapter 01: Introduction
Senior Data Specialist & Applied AI
I've always loved the work, the learning and the moment it clicks.
Either 100% dedicated or nothing. I enjoy the thinking, the building and the process of getting there.
A simplified solar system with a sun, Earth and three planets on thin orbital paths. Drag or use arrow keys to rotate it; press Home to reset.
When my curiosity gets triggered, I literally can't get away from it — and I've realized that's genuinely valuable to some people.
I've always loved doing my work and learning new things. Either 100% dedicated or nothing.
I enjoy tutoring and teaching, and I've always enjoyed building the gap between numbers, data and business to bring value into practice. I love the theory behind everything, but I know how important it is to ship.
I'm at my best when I understand the problem first, then find the solution that's actually needed — not the one that looks best in a slide.
Now
How I work
3 principles
Passion first
I'm either all in or not in at all. The work matters to me, and it shows in the details.
Think from scratch
I like starting with a blank page, questioning the obvious and understanding the problem before touching a model.
Enjoy the process
The thinking is part of the value. I like the theory, but shipping is what makes it real.
Beyond the screen
Motorsport, badminton, Brazilian Jiu-Jitsu, cycling, volleyball and chess. Off the clock I also build local AI setups, spend time in ComfyUI, and design and print parts in Autodesk Fusion. Curiosity doesn't stop at the tab.
Problem, approach and measured result. Open a case study to read the whole thing.
Stock Meta Ray-Ban glasses expose no camera or microphone API, so a sub-second assistant has to be built on top of a closed Messenger call — without exporting a single frame of a bystander's face.
PyTorchCUDAfaster-whisper
Social group dining requires coordinated scheduling, location filtering and party capacity without race conditions or manual RSVP tracking — and a browser is not a trust boundary.
Next.js 15 (App Router)React 19TypeScript
Cross-venue arbitrage lives for a few milliseconds: a nested-loop Python scan of 120+ pairs took 45 ms per cycle check, so every spread it found had already closed before the order could be signed.
Python 3.11NumPyNumba JIT
A coach reads a CGM export and a week of meal photos by hand. Joining the curve to the food that caused it is slow, subjective and impossible to reproduce — and a health record is the last place an LLM should be trusted to write freely.
Python 3Gemini 2.5 Flashgoogle-generativeai
A coaching roster means a daily message to every client. The naive shape is a script per client, a central recipient table, and a QR login every morning — one missed login silently skips the entire book of business.
Python 3Selenium 4Chrome headless
A Pi 4B has no GPU budget for a face tracker, a 92 MB offline recogniser and a 24 kHz voice stream at the same time. Run all three naively and the robot becomes a laggy speaker with a camera.
Python 3.11Raspberry Pi 4BOpenCV (YuNet)
A pack drop is decided in the first few hundred milliseconds. Sleeping until the sale opens loses that race to OS scheduler granularity, and every second spent re-finding a button is a pack someone else bought.
Python 3.9Selenium 4ChromeDriver
A game canvas, a wallet and an OAuth popup are three browser windows whose only shared state is whichever one has focus. The first version treated any failure as fatal and paid a full cold re-authentication for a single lost handle.
Python 3.9SeleniumChromeDriver
Every engagement field sits behind a different DOM path, a virtualised likes dialog and a localized UI. A scraper that fails on a missing geolocation does not report the gap — it drops the row and quietly biases the analysis.
Python 3.9Seleniumpandas
The problem, what I built and what changed because of it.
Algorithmic pricing strategies and ML elasticity models across commercial franchise networks.
From app telemetry to €2B+ in annual digital sales.
High-throughput Databricks data infrastructure and statistical experimentation.
The AI, data, experimentation and backend tooling behind the case studies above.
AI & Deep Learning
Data Engineering & Scale
Experimentation & Math
Backend & Software
Physics at LMU Munich, then an M.Sc. in Applied Artificial Intelligence completed alongside my full-time role at McDonald's. The master's thesis was graded 1.0 [best].
South Westphalia University of Applied Sciences (Fachhochschule Südwestfalen) · Iserlohn, Germany
Oct 2023 — Jan 2026 · Master's thesis: 1.0 [best] · Mit Auszeichnung bestanden
Ludwig Maximilian University of Munich (LMU) · Munich, Germany
Oct 2018 — Jun 2021 · Thesis advisor: P.D. Dr. R. Saglia
Enterprise strategy with C-level executives and franchise boards.
M.Sc. AI completed entirely in English; technical publications and specs.
Native family and bilingual proficiency.
Teaching & tutoring
I enjoy explaining a hard idea until it clicks. Tutoring and teaching keep my thinking honest.
I'm open to senior applied AI, machine learning and lead data science roles from January 2027. Email is the fastest route.
