Scientific Assistant (full-time) — FHNW, Windisch
2024 — present
- Take substantial technical responsibility for applied AI/ML projects with industry partners — requirements clarification, exploratory analysis, feasibility assessment, model development, on-site workshops, technical alignment with partner IT, and pilot planning.
- Building a structured document-extraction pipeline for PDFs and scanned documents using vision-language and LLM APIs: customer-specific JSON schemas, zero-shot AI pre-labeling, Langfuse tracing and prompt versioning, with a held-out evaluation set and F1-based protocol to compare prompt/model variants.
- Conducted a feasibility study for real-time AI guidance on Swiss German call-center audio, integrating speech-to-text and LLM-based recommendations; built the guidance prototype on Azure AI and Google Vertex AI, assessing model quality, latency, and cost.
- Built an optimization framework for wastewater-treatment plants achieving 40× faster optimization (<3s for a 6.7k-step, 5-component system) — published in Aqua & Gas 4/2026.
- Develop predictive and control models on industrial time-series data (MAPE < 10%; scikit-learn/AutoGluon, Hydra, Weights & Biases).
- Trained ML models on 300k pollen hologram pairs (80 GB) with PyTorch Lightning, DVC, and SLURM, with reproducible data and model versioning.
Lecturer (part-time) — ABB-Technikerschule, Baden
2024 — present
- Develop and teach a Data Science & AI course for Gebäudeautomation HF / Energie- und Umwelttechnik HF.
CTO & Co-Founder (part-time) — Studyflash, remote
2023 — 2024
- Built and operated a production LLM-backed learning platform serving 20k+ users, reaching CHF 20k monthly revenue and 3M+ flashcards; developed document-ingestion and flashcard-generation services using commercial LLM APIs.
- Established QA for LLM-generated content using blind human evaluations and regression tests; iterated on extraction and prompt logic based on observed failure modes, including validation and retries for invalid outputs.
- Architected the migration from a no-code prototype to a scalable backend (Vue.js, Supabase, Docker, Stripe), building core services for auth, billing, and document processing.
- Led a three-person development team, planned and prioritized engineering work, and participated in pitching the company to investors, including a successful pitch on Die Höhle der Löwen Schweiz.
Data Analyst (part-time) — XLnow, Zuchwil
2021 — 2023
- Implemented a Power BI data model for time-tracking/project metrics, improving reporting speed and data consistency.
Apprentice Software Engineer (full-time) — Corix AG, Biberist
2017 — 2021
- Delivered a full client project (~1 person-year) end-to-end while coordinating a two-person team.
- Developed core ERP components in Java 8 with SQL Server, introducing reporting, automation, and UI frameworks.
- Applied agile practices (Scrum, Kanban) and tooling (Git, JIRA, Ant) for reliable releases and continuous integration.
FHNW — BSc in Data Science, Windisch
2021 — 2025
- Graduated Feb 2025 with distinction; overall grade 5.6/6.0.
- Bachelor thesis: satellite-based classification of crop types and cultivation methods (grade 6.0).
Corix AG / BBZ Solothurn — Informatiker EFZ, Applikationsentwicklung
2017 — 2021
- Top of class (GPA 5.7); final apprenticeship project with distinction (6.0).
- Completed Berufsmatura (Technik, Architektur, Life Sciences) alongside the apprenticeship.
Crop classification with temporal vision transformers — FHNW
2024 · github.com/Satellite-Based-Crop-Classification/messis
- 82.9% field-level accuracy on Swiss Sentinel-2 data by fine-tuning the geospatial foundation model Prithvi and designing the Messis architecture for multi-temporal inputs.
- Reproducible ML pipelines (PyTorch Lightning, DVC, Hydra) and a novel stratified data-splitting method for balanced cross-validation.
Smart classroom air quality optimization — FHNW
2023 · github.com/Smart-Classroom-Challenge/Abgabe
- Distributed IoT system (Arduino, Raspberry Pi, MQTT, Django, TimescaleDB) processing >100k readings/week to stabilize classroom CO₂; prototype rolled out to three classrooms.
Real-time notification service — Kaschuso Systems
2020 · github.com/KaschusoSystems/kebab
- Full-stack notification system for the cantonal ERP kaschuso.so.ch; secure Node.js backend (AES-256, JWT, diff-based polling) with automated Jest/Cypress testing.
Member — Swiss Study Foundation
2022 — 2025
- Selected for a national excellence program; extracurricular seminars and activities (NMUN NY, communication and detecting manipulation, etc.).
Youth work — CEVI, Zuchwil
2016 — present
- Lead J+S camps with 30 leaders and 40 participants; mentoring the team.
- Board treasurer; oversee financial operations and fundraising, co-organize events and a weekly group activity.