Reflections on navigating interdisciplinary research across three stages:

• Self-Adaptive Service Discovery: Service management in large and dynamic environmets
• The Auto AI Project: Deploying ML in real world systems
• The aICU Project: Deploying ML in the ICU with medical practitioners

• Systems Engineer. Universidad de Nariño, Pasto, Colombia
• MSc Systems Engineering and Computer Science. Universidad de los Andes, Bogotá, Colombia
• PhD in Computer Science. Trinity College Dublin, Ireland
• Assistant Research Professor, University of Cambridge

The problem: large environments require autonomous service-oriented architectures.

The goal: enable self-adaptive service discovery in dynamic urban environments. Service architectures should reorganise themselves in response to changing urban conditions.

I tried to use a RL-based approach which did not work as expected in realistic conditions mainy because:

• The real world was far more complex than the models assumed
• My own knowledge of RL and its limitations was still developing

This was a negative result, but it became part of my PhD thesis. Negative results are not failures, they are an honest account of what we tried and what we learned.

The experiments started to work when I introduced strong assumptions, simplifying the environment until it no longer reflected the real world.

RL worked, but only in a very limited and controlled setting.

ML can work in isolation, but deploying it in real-world environments is a fundamentally different challenge.

The experiments started to work when I introduced strong assumptions, simplifying the environment until it no longer reflected the real world.

RL worked, but only in a very limited and controlled setting.

ML can work in isolation, but deploying it in real-world environments is a fundamentally different challenge.

The insight from my PhD led me to a postdoc at the University of Cambridge, joining the AutoAI project in the ML@CL research group.

The problem: deploying ML models in the real world is challenging and they do not work as they do in the lab settings.

The goal: understand and address the challenges of deploying ML in the real world.

The project asked for a systems perspective, which aligned with my training as a systems engineer.

This was my first real interdisciplinary experience.

I started working with ML experts who used a very different vocabulary and had a model-centric view of the problem. The research was much more technical than what I was used to.

Challenges emerged quickly:

• Understanding their language and technical framing
• Making the systems perspective understood
• Recognising that both views were valid

My first instinct was to catch up with the ML experts.

I realised that it was not feasible.

What actually helped:

• Catch up enough to have informed conversations
• The problem as the main driver of our work
• Different views on the problem generate better informed solutions
• Value the systems perspective
• Giving time to the process of mutual understanding

The aICU project: deploying ML systems in the Intensive Care Unit to support practitioners in their daily activities.

A collaboration with medical doctors from the Karolinska Institute.

The team: doctors, data scientists, ML experts, and HCI experts. A much more diverse interdisciplinary setting.

In this case, catching up with the medical doctors is clearly not an option.

What is working instead:

• Finding a common vocabulary
• Being open to explore unfamiliar areas
• Making your point when needed
• Understanding different points of view
• Focusing on the problem first as the common denominator

Teamwork skills that helped across both experiences:

• Make the effort to explore unfamiliar areas and develop common understanding
• Valuing your own expertise
• Making your point when needed
• Giving time to the process

Teamwork skills that helped across both experiences:

• Make the effort to explore unfamiliar areas and develop common understanding
• Valuing your own expertise
• Making your point when needed
• Giving time to the process

Two systems engineering principles that underpin effective interdisciplinary work:

• Problem-first: The problem at hand is the common denominator across disciplines
• Systems views: The same problem can be seen from different perspectives. That is not a conflict, it is a strength

The teamwork skills are developed by every individual through experience.

But the systems engineering way of thinking, problem-first and systems views, is something we could all develop and use, regardless of discipline.