Reflections from a SAFEHR Intern

Luna Toma was supervised by SAFEHR’s Dr Steve Harris for her research internship in UCL’s Institute of Health Informatics. Her reflections provide insight into the exciting work happening at SAFEHR.

Reflections on My Research Internship at UCL’s Institute of Health Informatics

By Luna Toma

From February 10 to March 21, 2025, I had the privilege of joining the Health Algorithm Lab at the Institute of Health Informatics (IHI), University College London (UCL) for a six-week research internship.

Motivation

Coming from my fifth year of medical school in Bonn, Germany, where I am developing a machine learning model for my doctoral thesis, this internship provided an opportunity to experience a different research environment and explore how AI is transforming healthcare across countries.

In my doctoral research, I work under the guidance of a professor, focusing on using histopathological image analysis to assist physicians in diagnosing a specific eye lesion. While this setup offers personalised mentorship, it lacks the collaborative exchange that comes with being part of a structured lab team. Additionally, in Germany, my role is limited to the medical aspect of the research, as the algorithm development is handled entirely by computer scientists. I wanted to gain a deeper understanding of the algorithmic side of AI—how models are built, trained, and optimised—so that I could bridge the gap between clinical understanding and computational methods.

London’s diverse population makes it an especially compelling environment for AI research in healthcare, as it provides a more representative study population that helps mitigate biases. Moreover, within academic and clinical research settings, there is a commitment to inclusion. London-based AI projects can therefore develop more robust healthcare solutions with global applications.

Onboarding

After my acceptance, I had the opportunity to meet Prof. Shallcross and Dr Steve Harris online to discuss the details of my internship. However, before arriving in London, both Dr Harris and his team, as well as myself, had to navigate several bureaucratic challenges to facilitate my stay.

Upon arrival, I had to complete multiple training sessions, including fire safety, EPIC training, and many e-learning modules due to my interaction with patient data. These formalities took up a significant portion of my short stay, but they were fascinating in themselves—a level of thorough induction I wasn’t used to in Germany.

At the same time, Steve and his team helped set up all the necessary technical accounts for collaboration: Slack, EPIC, GitHub, Jupyter, and Elasticsearch. This period of logistical setup, though time-consuming, underscored the importance of data security and research infrastructure. One of the most critical aspects of working with AI in healthcare is understanding Trusted Research Environments (TREs). These are secure platforms designed to store and analyse sensitive patient data while ensuring compliance with stringent privacy regulations.

Research Environment

One of the most valuable aspects of this internship was the opportunity to interact with PhD students, researchers, and experts in AI and healthcare. During lab meetings, I got to know the team and learned how research is presented and discussed. Following Steve’s suggestion, I scheduled one-on-one meetings with PhD students to discuss their projects, challenges in the research world, and their academic journeys. These candid conversations provided invaluable insights, helping me understand the realities of research, career pathways, and the mindset needed to navigate academia.

Among various group meetings, the OneLondon Workshop stood out, where I met leaders and experts in AI-driven healthcare solutions. The brainstorming sessions inspired me as we discussed AI policy, data infrastructure, tool development across London, funding challenges and international research commercialisation.

Research Project

My project at UCL focused on predicting inpatient falls using the MedCAT tool, which extracts unstructured information from electronic health records. My task was to create filters in Elasticsearch to eliminate non-inpatient falls and then integrate these filters into the code. However, before that, we needed to establish a connection between Python and Elasticsearch—a process that involved a lot of trial and error.

As a medical student, I quickly realised that if you want to work in AI-driven healthcare research, there’s a steep learning curve—you have to teach yourself a lot from scratch, as these skills are not part of a standard medical curriculum. This challenge, however, is not unique to AI; it’s similar to what medical students face when working in wet lab research, where they need to acquire skills outside of traditional clinical training.

Reality of Short-Term Research Stays: A Shift in Perspective

Before arriving, I had unrealistic expectations about what I would accomplish in six weeks. I thought I would contribute significantly to a single project and leave with a mastery of a specific skill. However, research doesn’t work that way—especially not in such a short time frame. Initially, this realisation frustrated me, but discussions with PhD students and a conversation with Dr Steve Harris helped me shift my perspective. I learned that research is not just about coding and analysis—it involves securing ethical and clinical approvals, applying for grants, and waiting on contributions from multiple stakeholders.

Understanding these realities didn’t discourage me from research; instead, it helped me develop patience and resilience. I became more forgiving of myself when things didn’t work perfectly the first time, which allowed me to enjoy the learning process rather than fixate on rigid goals. This, I believe, is a valuable lesson for both research and life in general.

Learnings about AI in healthcare

One of the biggest challenges in interdisciplinary work is communication. It can be difficult to gauge the other person’s level of understanding. Despite this, for AI research in healthcare to be truly effective, both fields must work together seamlessly, bridging the gap between medical expertise and computational power. The best results come when teams actively recognise each other’s strengths, find common ground in problem-solving approaches, and develop a shared language to navigate the complexities of AI in medicine.

A particularly thought-provoking idea that stood out to me at the OneLondon Conference was the concept of applying entrepreneurial thinking to research—viewing AI development not merely as an academic endeavour, but as a process more akin to factories rather than libraries when considering commercialisation. This perspective underscores the potential for AI-driven medical solutions to be both scalable and financially sustainable, extending their impact beyond the research setting.

The benefits of AI research—whether in terms of improved patient outcomes or financial returns—should ultimately flow back into patient care. However, this shift toward commercialisation also necessitates careful ethical and legal considerations, particularly regarding data privacy. Some parameters within AI algorithms could, in theory, be traced back to patient data, raising serious concerns about confidentiality and regulatory compliance. Issues of data ownership and profit-driven research demand a balanced approach that aligns innovation with public interest while protecting healthcare equity and accessibility.

During my final lab meeting, I had the opportunity to present a comparative analysis of digital healthcare in Germany and the UK. Preparing and delivering this presentation not only enhanced my knowledge of international healthcare systems but also strengthened essential research skills, such as presenting clearly and engaging in academic discussions.

Summary and Takeaways

This internship was much more than just an opportunity to work on a project—it was a transformative learning experience. Even writing this blog post has been a valuable exercise in reflection, helping me process everything I’ve learned and appreciate how much I’ve grown in just six weeks.

This experience has reinforced my interest in AI-driven medical research, and I look forward to applying these insights to my doctoral work back in Germany. I am deeply grateful to Prof. Shallcross, Dr Steve Harris, Dr Claire Black, Kawsar Noor, and the entire team at IHI for their support and mentorship.

For anyone considering a short research internship: Go in with an open mind, embrace the unexpected challenges, and focus on learning rather than just producing results. You might be surprised at just how much you gain from the experience.