An article by Cohort 5 students Oluwadara Adedeji (UCD), Abigail Naa Amankwaa Abeo (DCU), and Thabang Isaka (TU Dublin)
Every year, I get this wrong. A few weeks before the programme begins, I pore over the schedule and carefully mark the talks that align most closely with my own research. Naturally, I assume those will be the highlights. The talks are excellent, I make notes, and I head home convinced I know what I’ll remember most.
But a week later, it’s never one of the talks I had marked that stays with me. It’s always something unexpected. Last year, it was the elevator pitch. The year before, the debate.
It happened again this June.
Over three days at the ML-Labs Summer School, jointly hosted with the Una Europa Summer School, I found myself drawn in by ideas I hadn’t expected to encounter. Abigail and Thabang are here to tell that story with me.
So, let’s start with day one.
Day 1
The morning had pastries, coffee, and a table of merch. There’s always a little excitement around the merchandise table. A totebag, a T-shirt—small souvenirs that have somehow become an exciting part of the experience. We got ours, found our seats, and Associate Professor Georgiana Ifrim opened the Summer School in a stunning green blazer, welcomed us and took us through the programme for the three days ahead.
Welcome and Opening Address by Prof. Georgiana Ifrim
In her opening remarks, she reminded us that the Summer School is more than machine learning. The technical skills, she said, are what get you through the door. But it is the connections you make, the social capital you build, that keep opening doors long after that first one. She kept returning to the idea of ML-Labs as a community, one that thrives when its people keep supporting and uplifting each other well beyond the PhD itself.
Thereafter, we had the first talk of the day which was an industry keynote delivered by Dr Binh Thanh Le on “What I thought was True before Joining Google”. You would expect a company like Google to run almost entirely on technical brilliance. Binh’s take was rather interesting. Talent, he said, is maybe 90% of it. The other 10% is luck, timing, and the people you happen to know. He offered it as an encouragement, though it cuts both ways, and it landed straight on Georgiana’s opening: the skills get you in, the relationships shape what happens next.
He built the rest of the talk around five myths about working there. The first is the one everyone believes, that every Googler is a human calculator, untangling algorithms in their head and returning perfect solutions on the spot. The reality he described is far more ordinary. People reread the documentation, ask each other questions, and solve things together. One story stuck with me. A colleague looked like he wasn’t working, spending time playing games with the team, but yet insisted he was still working. And he really was. The trust and the conversations built over those games were really worth it.
The second myth is that Google is a gladiator pit, with everyone racing to outshine everyone else. What he described instead was Googliness: a culture where helping someone else succeed counts as your success, and collaboration matters more than beating the person next to you.
The third is about work-life balance, and the fear that a job like this eats your whole life away. He painted something more flexible. People have lives, the schedules bend to fit them, and most split the week now between a few days in the office and the rest working wherever they want.
Industry Keynote by Dr. Binh Thanh Le, Google
He also talked about how much room there is to move around inside the company. After the first year you can apply to a different team, and about a fifth of your time might already be spent working with other teams or in other locations. Even the building is built for it. The game rooms, the gyms, the pools, the shared spaces, all of it is there to get people interacting rather than just looking good in a brochure.
The fourth myth follows from that: the fear that you join Google and spend years stuck on one feature, going nowhere. He said the opposite is true. People move between teams, there are tech talks and workshops most weeks, and mentoring is structured to actually push your learning.
The last myth is the famous one, the free food. He did not pretend the perks are not real; the meals, the cuisines, all of it. But he was firm that none of it is the point. What makes the job worth doing is the scale, solving problems that reach billions of people. The food is nice. The reach is the reason.
He closed with a few myths about the CV itself, for anyone hoping to get in. You do not need five pages listing every project you have ever touched. Keep it short, keep it focused, and let some of your own personality through. Then the practical bit: make problem-solving a daily habit, on LeetCode, and get your SQL sharp, especially for the data science and ML roles.
After that we had a break, and then a surprisingly competitive game of Bingo broke out among us.
Bingo Game
The next talk did what the best of them always do. It took a word I had heard plenty of times and never once stopped to actually understand: Serendipity. Finding something valuable whilst you were looking for something else entirely. Dr Dorota Glowacka’s talk, “Serendipity in Search and Recommendation”, made me feel like I was hearing it for the first time. The Una Europa group joined us for this one.
She started with where the word comes from, which I loved. Serendipity traces back to Serendip, the old Persian name for Sri Lanka, and to the tale of the Three Princes of Serendip, who kept stumbling on things they were never looking for whilst chasing something else.
Dr. Dorota Glowacka’s talk on Serendipity in Search and Recommendation
Dr. Glowacka connected this idea to modern search and recommendation systems, highlighting the delicate balance between showing users content they are likely to enjoy based on past interactions and introducing them to unexpected yet valuable discoveries. She argued that recommendation systems should strive not only for higher engagement or better ratings, but also for meaningful and sensible recommendations that broaden users’ horizons.
She ended with a thought-provoking question: in the age of large language models and AI, where interactions are increasingly driven by concise prompts and direct answers, how do we preserve room for serendipity?
I walked into the session unsure of what serendipity really meant. I left not only with a clear understanding of the concept, but also with a new appreciation of its relevance to machine learning and my own research. More importantly, it inspired me to venture beyond my usual reading list of AI and machine learning books. After all, the next great idea may well be sitting where I least expect to find it.
Day 2
Day two began with another exciting lineup of talks, picking up seamlessly from the momentum of the previous day. The programme featured sessions on Open Science, AI Governance, Machine Learning for Biodata, and an update from Georgiana about the activities of ML-Labs.
The first talk, AI and Open Source, was delivered by Claire Dillon, who drew on decades of experience in IT and open-source software. She opened with a question that had been on my mind for some time: In the age of AI, where code is increasingly democratised, what is the real competitive moat?
One of the most surprising takeaways was just how deeply open source is woven into modern technology. An estimated 96% of software contains open-source components, and even proprietary software is often built on open-source foundations—a practice known as innersource. In other words, the democratisation of technology did not begin with AI; open source has quietly shaped the software industry for decades and is now extending its influence far beyond the technical community.
AI and Open Source Talk delivered by Claire Dillon
Claire then dispelled several common myths about open source: that it is merely “free” software, inherently less secure, maintained only by volunteers, incompatible with commercialisation, or capable of maintaining itself. In each case, the reality proved far more nuanced and, in many instances, the opposite of what people commonly assume.
The discussion then shifted to Open Source AI, where she expanded the traditional definition of open source to include not only open code, but also open model weights and open datasets. She argued that many models marketed as “open source” fall short of that standard because they withhold the training data, limiting transparency and reproducibility. Encouragingly, she noted that open-source AI models are steadily closing the performance gap with their closed-source counterparts.
She concluded by exploring the future of open source in what she described as the post-AI-subsidy era. Among the emerging challenges were rising cybersecurity threats, increasing token costs, and a widespread lack of understanding of how to sustainably govern open-source ecosystems. Yet she also highlighted a significant opportunity for Ireland to become a leader in this space. By the end of the talk, the question she posed at the beginning had found its answer. In a world where code is increasingly accessible, competitive advantage lies less in exclusive access to technology and more in delivering exceptional products, services, and access to markets. Technology may become commoditised, but execution remains the true moat.
Talk by Kieran Towey
The next talk, delivered by Kieran Towey, explored AI Risk Management and Operations, a topic that sits at the heart of responsible AI deployment. He began by unpacking what AI governance really means: not just rules and regulations, but the guardrails, tools, data practices, and processes that ensure AI systems are developed and deployed responsibly.
One example that stood out was the wave of AI models developed during the COVID-19 pandemic. While many showed promise in controlled settings, most proved unfit for real-world deployment because they learned spurious patterns rather than meaningful clinical signals. It was a powerful reminder that strong model performance does not always translate into real-world value.
Another insight that surprised me was his description of life in industry. Contrary to the perception that AI jobs revolve almost entirely around building models, he explained that only a small fraction of the work is spent on model development. The majority is devoted to documentation, governance, compliance, testing, and the operational processes needed to deploy AI systems safely and reliably.
Kieran also highlighted a sobering reality: despite the excitement surrounding AI, many organisations struggle to translate new technologies into tangible business value. He noted that an estimated 95% of generative AI projects fail to deliver measurable outcomes. The challenge, therefore, is not simply building AI systems, but governing them in a way that balances innovation, risk management, and business impact.
He concluded by outlining the diverse risks associated with AI deployment—including technical, ethical, legal, operational, and societal risks—and introduced several established frameworks for managing them. His closing message resonated strongly with me: governance should not be viewed as an obstacle to innovation, but as an enabler of trustworthy AI. By identifying risks early and being transparent about them, developers can build systems that are not only powerful, but also safe, reliable, and worthy of public trust.
Georgiana returned to update the community about the activities and giant strides of ML Labs.
Alumni Survey update by Prof. Georgiana Ifrim
She shared an update on the journeys of the programme’s alumni. Among the 26 graduates surveyed one year after completing the programme, 65% had moved into industry while 35% pursued careers in academia. Notably, 65% remained in Ireland, with the rest building careers abroad—a testament to the programme’s strong international reach and the global demand for its graduates.
It was particularly encouraging to learn that every respondent was gainfully employed. Many had secured highly competitive salaries, with 34.6% earning €80,000 or more, including 26.9% earning above €100,000 annually. Their roles spanned senior engineering and software engineering positions, applied scientist roles in industry, as well as postdoctoral and academic appointments.
Georgiana encouraged us not to undersell ourselves, reminding us to be ambitious and confident enough to apply for senior roles. The alumni, in turn, credited much of their success to the programme’s blend of industry experience through internships, rigorous machine learning training, and the soft skills that prepared them to thrive beyond the classroom.
Luca Costabello presenting
The final talk of the day was delivered by Dr. Luca Costabello, a Research Scientist at Accenture Labs, who spoke on machine learning for knowledge discovery in biodata and generative computational drug discovery. He showcased how knowledge graphs and AI can uncover novel gene-disease relationships, accelerating scientific discovery and opening new possibilities for drug development.
Interestingly, he returned to the theme of serendipity that had featured so prominently in the previous day’s discussions. In drug discovery, some of the most important breakthroughs emerge not from finding exactly what was set out for, but from uncovering unexpected connections along the way.
While the research itself was fascinating, what captivated me most was Dr. Costabello’s journey. Despite coming from a technical background, he had successfully transitioned into the highly specialised world of computational biology. He generously shared that this transformation did not happen overnight, but was built through persistent reading, curiosity, and a willingness to continually learn. It was a reassuring reminder that expertise is often cultivated, not inherited.
He also encouraged us to consider opportunities in the pharmaceutical industry, where AI talent is increasingly in demand. He concluded with a reflection that extended far beyond machine learning or biology. Regardless of the topic of our PhDs, he said, the doctoral journey equips us with something even more valuable: the ability to navigate uncertainty, tackle complex problems with persistence, work independently, and see challenging ideas through to completion. These, he reminded us, are the skills that endure long after the PhD is over.
Day 3
The final day of the ML-Labs Summer School began with a panel discussion featuring two alumni, Dr. Di Meng and Dr. Mark Germaine, who returned to share their experiences during and after their PhDs. Moderated by Thabang Isaka, the session offered an honest and reassuring glimpse into the realities of doctoral life.
Alumni Panel Discussion with Dr. Di Meng and Dr. Mark Germaine, moderated by Thabang Isaka
The discussion opened with a topic that resonates with almost every PhD student: rejection. Di shared that she did not publish her first paper until her third year and admitted that it was difficult to watch many of her peers publish while she had none. Rather than chasing publications, she encouraged us to focus on learning deeply and producing high-quality research. The publications, she assured us, would follow.
Mark echoed a similar experience. Coming from a non-computer science background, he struggled with imposter syndrome during his first year. Progress was slow, and it took nearly a year before he had the data needed to begin his research. Like Di, most of his publications came towards the latter stages of his PhD. Their stories were a timely reminder that every research journey unfolds at its own pace.
The conversation then shifted to managing relationships with supervisors. Mark reflected on the challenge of transitioning to a new supervisor midway through his PhD, a change that ultimately proved positive. Di, meanwhile, spoke about navigating the dynamics of working within a multidisciplinary research team, highlighting the importance of communication, openness, and collaboration.
Life beyond research also featured prominently in the discussion. Having started her PhD during the COVID-19 pandemic, Di found balance through exercise, running, and swimming. Those moments away from her desk helped her recharge and return with renewed focus. Her message was simple but powerful: a successful PhD is sustained not only by hard work, but also by making time to rest and care for yourself.
When asked about thesis writing, both speakers agreed on one principle: start early. Mark recommended setting firm deadlines and breaking the thesis into manageable chapter-by-chapter milestones, noting that the literature review often takes longer than expected and is best tackled early. Di reinforced this advice, encouraging us to begin writing long before the submission deadline looms.
The panel concluded with a discussion of career paths after the PhD. Di shared that she eventually realised research is not confined to universities—companies conduct cutting-edge research too, and academia is only one of many possible destinations. Mark highlighted the importance of building strong professional networks when transitioning into industry and praised the ML-Labs industry placement programme as an invaluable stepping stone.
The session ended with a lively Q&A, where the panelists answered questions on recruitment, technical interviews, building confidence, and the growing role of AI tools in software development. It was an honest, practical, and deeply enriching conversation that left many of us feeling both reassured and inspired.
Social Programme: Lunch & Farm visit at Airfield Estate
The Summer School concluded with a visit to Airfield Estate, where we explored its working farm, beautiful gardens, and rich history dating back to 1894. It was a fitting way to end three days of learning, reflection, and connection.
Although it was bittersweet to know this would be the final edition of the ML-Labs Summer School, it was equally comforting to realise that its true legacy lives on—not in the event itself, but in the friendships forged, the lessons learned, and the countless lives it has shaped over the years. Some experiences end, but their impact endures.
Summer School Group photos