Building Trustworthy ML in Healthcare: Inside “Adrian Schiegl, Head of Data Science von XUND” on culture, hiring, and engineering rigor

A technology-led team that treats responsibility as a first-class requirement

When a company claims to be technology-driven, we at DevJobs.at look for the evidence: how the team is structured, how work is executed, and how people are hired, onboarded, and empowered to grow. In the session “Adrian Schiegl, Head of Data Science von XUND,” we heard a concise and grounded perspective on all of the above.

XUND concentrates its engineering and data capabilities across two hubs—Budapest and Vienna. “Right now we are around 15 people at XUND working in software development and data science. That makes up about 50 percent of our employees. So we are a very technology-driven startup.” The data and ML core sits in Vienna: a small team today with an ambitious plan to scale. “We are two to three people and plan to grow to about ten by the end of the year.”

What this unit does is focused and explicit: “Our team, although it is a data science team, is actually a machine learning team. We work exclusively on machine learning and in the medical domain.” There is a clear NLP emphasis—effectively deep learning—implemented in Python with PyTorch. And the bridge between research and product is built within a regulated environment: XUND operates as a medical device manufacturer, with the quality management, data protection, and trust requirements that implies.

Structure and location: How XUND organizes engineering

The team composition is crisp: In Vienna, data scientists, data engineers, and machine learning engineers form the core. “In my team there are exclusively data scientists, data engineers, machine learning engineers. Of course, as the team grows, more roles will be possible.” The development team in Budapest provides a template of rigorous software practices that the ML team adopts deliberately: GitLab for source control, continuous integration and regression testing, and an emphasis on professionalism, reproducibility, and collaboration.

This split offers a real advantage. Vienna concentrates ML depth in NLP and deep learning, while Budapest sets the delivery standard from classical software engineering. Together, they shape a culture that blends research with product readiness—without compromising on quality or safety.

Engineering culture: Research, product, and proven practices

XUND combines a research-oriented mindset with disciplined engineering. The ML work doesn’t happen in isolation—or only on laptops. “We also largely do not work on our laptops because we don’t have GPUs powerful enough. That’s why we work a lot with cloud providers who give us these GPUs and 100-core CPUs.” That reflects realistic production constraints: large NLP models, deep-learning workloads, and demanding training runs need strong, scalable infrastructure.

Technologically, the choices are clear: Python as the language, PyTorch as the deep learning framework, and NLP as the primary application area. Working here means working close to state of the art—measured not only by model metrics but by the standards of medical practice, data protection, and trustworthiness.

At the same time, the team imports software engineering discipline from Budapest: “We use GitLab as a source control system. We also use continuous integration and regression tests and generally try to bring many principles from software development into data science, because those methods have proven themselves over time.” That line matters. It signals that data science at XUND is treated as a rigorous engineering activity, with the same expectations for code quality, testing, and maintainability.

Research in a regulated setting: Ambition with accountability

“There is no shortage of technical challenges.” That’s not a throwaway line when you build ML for healthcare. XUND collaborates with universities and medical institutions, aims to publish, and targets performance close to state-of-the-art models. The crucial add-on is the medical context: “Because we operate in medicine, we also need to respect patient privacy, develop models that are trustworthy—so you can release them into the real world with a clear conscience—ensure robustness, and include human input when needed. Sometimes that’s even the primary requirement.”

This is where the real job begins: not just accuracy, but accountability. Trustworthy AI in healthcare demands data protection, stability, robustness, and—where necessary—a strong human-in-the-loop. For ML professionals, that’s an attractive challenge: cutting-edge modeling with tangible impact, within clear ethical and regulatory guardrails.

Hiring philosophy: Fair, focused, respectful

The hiring approach stood out. The data science team is “involved in every step of recruiting”—from drafting requirements to screening CVs—and the founders are actively engaged. The principle behind it is noteworthy: “It’s a personal concern not to waste candidates’ time.” That’s reflected in a streamlined process.

The three-step interview flow

1. Introductory interview: A get-to-know-you session with Adrian Schiegl and a founder, aligning expectations. Crucial question: Will the candidate feel at home in the team and role?
2. Technical interview: Instead of puzzles or IQ tests, candidates tackle a practical, work-like exercise. “I’m not interested in doing intelligence tests, because I don’t think they are meaningful for later job performance.” The task is relevant—and time-bounded—to respect candidates’ effort.
3. Final conversation: Agreement and terms. The data science team steps out at this stage—by design—once the technical decision is made and the conversation moves to formalities.

This is the kind of process many engineers hope for: concise, relevant, and respectful. It also signals how XUND collaborates day to day—focused on real work, anchored in team ownership, and grounded in mutual respect.

Onboarding: Quality, vision, and hands-on enablement

Onboarding is short but sharply prioritized—fitting for a growing team in a regulated environment.

• Quality management: “At the beginning there is an introduction to our quality management system, because we are a medical device manufacturer.” Everyone must understand the standards to apply them in daily work.
• Vision and ethics: Founders lead a session discussing the company’s vision and ethical principles. From day one, the why is explicit.
• Technical onboarding: “We introduce the new hire to the tools, show where our documentation is, and identify opportunities to update the documentation.” Expectations are calibrated: “We don’t expect people to be super productive from day one or week one. We give people the opportunity to learn on the job and acquire new skills.”

This triad—standards, purpose, and tools—creates immediate orientation and communicates a learning mindset: ramp-up is intentional, not accidental.

Tech stack and execution: Python, PyTorch, cloud—and discipline

XUND’s technology choices are intentional and uncluttered:

• Python as the foundation for data science and ML
• PyTorch for deep learning, with an NLP focus
• Cloud resources for GPU- and CPU-intensive workloads (“100-core CPUs”) instead of local workarounds
• GitLab for version control
• Continuous integration and regression testing to ensure stability and reproducibility
• Proven software engineering principles adopted into data science workflows

For candidates, this is encouraging. You won’t be held back by legacy tooling or ad-hoc processes. Instead, you’ll work in a modern ML environment that treats production-grade experimentation and reproducible research as the norm.

Collaborating with academia and clinics: Publish and productize

“We often work with universities and medical institutions within research projects, where we also try to publish and come as close as possible to state-of-the-art models.” For many ML profiles, that’s the sweet spot: scientific rigor and real-world data domains, combined with product requirements like robustness, privacy, and trustworthiness.

This opens meaningful paths within the team: those inclined toward model research can push on SOTA approaches; those who love making systems production-ready can focus on CI, regression testing, cloud execution, and similar; those who thrive at the domain bridge can translate clinical constraints into ML practice—up to and including settings where human input is the primary requirement.

Everyday culture in practice

The session highlighted several cultural pillars:

• Practice over puzzles: Technical ability is assessed through relevant, work-like tasks—not intelligence tests.
• Respect for time: Exercises are time-bounded; the interview process is three steps, end to end.
• Quality first: As a medical device manufacturer, standards are non-negotiable; onboarding reflects that.
• Learning mindset: No one is expected to be “super productive” on day one; on-the-job learning is encouraged.
• Research with responsibility: SOTA ambition meets privacy, robustness, and trustworthiness.
• Software discipline for data science: GitLab, CI, and regression tests are embedded in the workflow.

Altogether, this is a sustainable setup: ambitious yet grounded; fast without being frantic; growth-oriented within clear quality boundaries.

Why XUND is compelling for tech talent

From our vantage point, there are several strong reasons engineers and ML professionals will find XUND attractive:

• Impact with accountability: ML in healthcare with clear responsibility and tangible benefit.
• Modern ML practice: NLP, deep learning, PyTorch, and cloud GPUs—without makeshift solutions.
• Solid engineering backbone: Version control, CI, regression testing, and principle-driven development.
• Respectful candidate experience: A lean, relevant, time-conscious interview process.
• Learning culture: Calibrated ramp-up, with room to acquire new skills on the job.
• Academic and clinical collaboration: Research projects with universities and medical institutions, including opportunities to publish.
• Growth runway: A team scaling from a small core—creating space for ownership, responsibility, and new roles.

If you want to grow technically while doing work that clearly matters, this is a rare combination.

Profiles that fit well

Given the stack and the way the team operates, the following profiles are likely an excellent match:

• Data scientists, ML engineers, and data engineers with a strong Python background
• Deep learning experience (PyTorch), ideally with NLP
• Familiarity with GitLab, CI, and a testing culture
• Comfort using cloud resources for GPU/CPU-intensive work
• Interest in research and collaboration with academic/medical partners
• Sensitivity to privacy, trust, and robustness in production systems
• Openness to human-in-the-loop designs where appropriate

Importantly, not every skill must be present on day one. The team expects learning curves and supports skill-building on the job.

Scaling with intention: More roles, more ownership

With a planned tripling of the Vienna ML team to roughly ten people, natural role diversification will follow. “As the team grows, more roles will be possible.” For candidates, that means potential for leadership, architectural responsibility, methodological ownership, or interfaces to quality management. The key is that culture remains consistent: practical methods, respect for time and focus, and onboarding that ties together quality, vision, and tools.

Memorable lines from the session

A few quotes encapsulate XUND’s approach:

These aren’t just words—they’re built into processes and day-to-day practice.

Our takeaway

“Adrian Schiegl, Head of Data Science von XUND” offered a clear window into an ML team that treats responsibility as integral and engineering discipline as non-negotiable. The mix of research and product, the medical context, the learning posture, and a respectful hiring process create an environment that is demanding yet realistic. If you’re at home in NLP and deep learning, comfortable with Python and PyTorch, and motivated by cloud-enabled ML with CI and testing baked in, your work here will stretch you technically and matter in practice.

From team structure to tooling to onboarding, XUND shows how to make machine learning work in a regulated domain—with pragmatism, care, and a focus on what ultimately counts: trustworthy, robust models that help people in the real world.