Understand first, model second.
I spend the first days not in code but in conversations. With the people who'll actually use the system. The best prompt in the world is useless if the question behind it is wrong.
→ Interviews, not briefings
Fifteen years
between
domain & model.
I'm Mark — PhD in Cognitive Systems, software engineer, sometimes university lecturer. I translate between subject matter and engineering, between research consortia and startups. My craft is making AI understandable — without softening it.
§ 01 — Portrait
Who's writing here.
// Mark WernsdorferBerlin
vitae.txt
Dr. Mark Wernsdorfer
// PhD Cognitive Systems, Uni Bamberg
// lives & works in Berlin
// speaks DE · EN · FR
// between model and domain
// PhD Cognitive Systems, Uni Bamberg
// lives & works in Berlin
// speaks DE · EN · FR
// between model and domain
Half researcher, half engineer. I come out of foundational work — symbol grounding, consciousness, the weirder corners of cognition — and I've spent fifteen years building ML systems where correctness touches lives, newsrooms, or classrooms.
PhD at the Otto-Friedrich-Universität Bamberg on symbol grounding — on how artificial systems develop meaning. Awarded the Springer Best Paper at the 11th AGI Conference in Prague (2018). Since then: Key AI Architect at University Hospital Leipzig (AMPEL), Research Associate at FAU Erlangen-Nürnberg, and Lead AI Developer for The Bina School.
The practice swings between two poles: foundational inquiry into how meaning forms in artificial systems — and the concrete building of ML systems in places where correctness counts: clinical labs, editorial newsrooms, classroom software.
Machine Learning
LLMs · TensorFlow · PyTorch
Python · Java · C++ · JS
Clinical Decision Support
Symbol Grounding
Cognitive Science
EU-AI-Act
§ 02 — Stance
Three principles you can hold me to.
Honest about limits.
If AI is the wrong tool for your problem, I'll say so. Sometimes it's a database. Sometimes a form. Sometimes a hard decision no model can take for you. I don't sell what doesn't fit.
→ No is part of the job
Hand off, don't wall in.
Success means you don't need me any more. Every project ends with documentation, handover workshops, and a team that understands the code and can keep building. No lock-in, no retainer trap.
→ Mission-complete is the goal
§ 03 — Vita
The full track record.
2026
Fellow — Prototype Fund (round 19)
Open Knowledge Foundation · BMFTRLLMSecTest: open-source test suite covering the OWASP LLM Top 10. Pytest plugin with SARIF output, CI/CD templates, 100+ red-team prompts. Six-month funding with a second-stage extension applied for.
2025
Lead AI Developer — The Bina School
Cognia-accredited online school · 40+ countriesLeading AI development on the Precision Education platform: adaptive learning paths, real-time progress tracking, personalised curricula for ages 4 – 12.
2024
Award Finalist — Innovation in Journalism
Media Lab Bayern · 4D Digital Rights by Design AwardDoppelCheck named a finalist at the 4D Digital Rights by Design 2024 Award for browser-based tooling against disinformation.
2023 — 24
Fellow — Media Lab Bayern & Prototype Fund
BMBF · WPK Innovation FundDeveloped SpotTheBot and DoppelCheck — AI tools for detecting disinformation and AI-generated text, with support from the BMBF and the WPK Innovation Fund.
since 2021
Research Associate — FAU Erlangen-Nürnberg
GeoZentrum Nordbayern · KNW-Opt · META studyMachine learning for shallow geothermal systems: predictive modelling and optimisation, including peer-reviewed work on thermal-conductivity prediction.
2019 — 21
Key AI Architect — AMPEL, University Hospital Leipzig
Universitätsklinikum Leipzig · JMIR Medical InformaticsArchitect of a Clinical Decision Support System, deployed across 1,500+ hospital beds. Co-first author of the primary AMPEL publication in JMIR Medical Informatics (2021).
2018
Springer Best Paper — 11th AGI Conference, Prague
AGI 2018 · "How Failure Facilitates Success"Award for the dissertation work on symbol grounding. Published in Springer Lecture Notes in Artificial Intelligence.
2014 — 18
Data Scientist — Orpheus GmbH
Procurement analytics · VC-funded startup (now McKinsey Spendscape)Machine learning for procurement analytics: automated data extraction, AI-driven spend categorisation, predictive analytics.
2011 — 14
Doctoral Fellow — BayEFG
Otto-Friedrich-Universität Bamberg · Bavarian Elite SupportCompetitive doctoral fellowship for research on symbol grounding and cognitive systems.
§ 04 — Expertise
Fields I know deeply.
Medical & healthcare
Clinical documentation, rare diseases, guideline search. Experience with GDPR & MDR.
Public sector
Administrative processes, citizen dialogue, parliamentary documents. On-prem & sovereign-cloud setups.
Industry & SMEs
Quality assurance, service knowledge, tender assistance. Focus on integrable solutions, not replacement.
Research & academia
Grant proposals, consortium projects, scientific evaluation. BMBF, Horizon Europe, Interreg.
— What's not on the list: what I don't know, I say. —
§ 05 — Publications
Selected writings.
Peer-reviewed · selection
-
Journal · 2021
Wernsdorfer, M. et al. — "AMPEL: A Clinical Decision Support System for Real-Time Patient Monitoring."
JMIR Medical Informatics, co-first author. Read the paper → -
Conference · 2018
Wernsdorfer, M. — "How Failure Facilitates Success: Symbol Grounding Revisited."
11th AGI Conference, Prague · Springer Best Paper Award. -
Dissertation · 2019
Symbol Grounding — Foundations of conceptual understanding in artificial systems.
Otto-Friedrich-Universität Bamberg. Read the thesis →
Research & visibility
-
Research · FAU
Peer-reviewed work on ML-based modelling of thermal conductivity in shallow geothermal systems (KNW-Opt, META study). FAU CRIS profile →
-
Open source · 2023 – 24
DoppelCheck & SpotTheBot — open-source tools for detecting disinformation and AI-generated text. GitHub profile →
-
Installation · 2023
ARTeFact (Spiegelbox) — interactive AI installation at Globe Theatre Leipzig, on the ethical dimensions of AI empathy.
§ 06 — Let's talk
Sounds like someone you'd work with?
I keep 2 – 3 first-call slots open per week. 30 minutes, no obligation, in German or English. The outcome: an honest read on whether I can help — and if so, how.