PhD in Computer Science and Informatics · Engineer & Researcher
I’m a builder with good taste: I take ideas all the way to working systems, and I know what to leave out. I work on problems until they yield, read widely outside the engineering canon, and prefer to find my own errors first.
Machine learning & mechanistic interpretability
Pre-registered and ran a study of cross-layer dynamics in transformer factual recall across six language models from three families, entirely on consumer hardware. All code will be made public once the paper is published. Also ran a grokking (delayed generalisation) study on gesture-recognition networks trained on electromyography data.
Paper: link TBA
openDAQ by Dewesoft and Hottinger Brüel & Kjær
Helped shape the open source data acquisition SDK that enables cross-platform, multi-language, and DAQ hardware vendor interoperability. Implemented core C++ features, documentation, CI/CD, release, testing, Python bindings, and metrics dashboards (Grafana/InfluxDB on AWS).
University of Maribor, Faculty of Electrical Engineering and Computer Science, System Software Laboratory
Led research in electromyography and teaching. Built end-to-end signal analysis pipelines (MATLAB, R), machine learning models for biosignals (Python, TensorFlow, Keras), and taught computer vision (OpenCV). Supervised students and contributed to peer-reviewed work.
Most cited paper: On the reuse of motor unit filters in high density surface electromyograms recorded at different contraction levels
Bibliography: COBISS researcher profile
University of Maribor, Faculty of Electrical Engineering and Computer Science
Python Machine learning Electromyography Dataset
Myo armband electromyography dataset of various hand gestures for machine learning.
Python Machine learning Electromyography
Electromyography gesture classification on Myo armband data, plus a deep investigation into producing grokking on real physiological signal.
Java Python Machine learning Electromyography LEGO
LEGO Mindstorms EV3 TRACK3R control using the Myo armband and a neural network for gesture recognition.
Python Machine learning LLM RAG
Local RAG pipeline over Slovenian documents — Ollama + ChromaDB, zero frameworks. Every step explicit and readable.
Python Machine learning LLM
Where does an LLM actually belong in a web-scale extraction pipeline? A rules → classifier → LLM cascade over funding-round news, with a hard grounding gate and evals.
Python Computer vision
Computer vision: find sudoku in a photo, extract grid, recognise digits, solve, and overlay solution on the image.
Kotlin Android
Android app for logging and tracking body weight over time.
Python Live charts
Discord bot that pulls close prices for crypto and stocks and displays them in a plot over the last N days.
Python Monitoring
Monitor for new job postings on slo-tech.com/delo.
Python Audio
Converts a regular wav file into an 8d audio wav file.
JavaScript Fun
Real-time webcam feed converted to ASCII art, running entirely in the browser with zero dependencies.
JavaScript Fun
Interactive gallery of generative art algorithms drawn live with p5.js — flow fields, fractals, the Mandelbrot set, reaction-diffusion and more.
C++ Python C# SDK
openDAQ™ data acquisition SDK — cross-platform, multi-language DAQ interoperability.
