While Teaching AI to Detect Abuse, I Learned to Think Differently: A PhD Journey of Transformation

Tulika BOSE

LinkedIn, ThesesFr

Merci beaucoup pour votre témoignage

Could you tell us about your background and what motivated you to do a PhD?

I’m originally from India, and after my Master’s, I worked as an AI researcher in the industrial research lab of a multinational company in India. It was a fast-paced environment—research was closely tied to business goals, and priorities could shift rapidly. I learned a lot, but after a few years, I realized I wanted to dig deeper into AI in an environment where I could focus on fundamental problems and have the time and independence to pursue them rigorously. A PhD offered that space—the time and freedom to pursue problems rigorously, to experiment, to make mistakes, and to learn from them. It was also a way to gain exposure to academic research methods, collaborate with experts, and develop skills that are highly valued in both academia and industry. Additionally, in industrial research, having a PhD is highly valued—it opens doors to leadership roles, managing research teams, and shaping the direction of a research unit.

With the support of my company, I was granted unpaid leave and took the leap to pursue a PhD in France—an exciting next step in my research journey that felt like diving into a world of uncertainty, and which, looking back, stands out as one of the best decisions I’ve ever made.

Can you please briefly describe your PhD thesis work?

My PhD focused on improving AI systems for detecting abusive language online—a task that is more complicated than it might seem owing to the constantly evolving nature of social media. Models trained on one dataset often fail on new platforms, changing language, or low-resource scenarios.

I addressed the gap between benchmark performance and real-world use by developing methods that allow models to transfer learning across datasets, even when data looks quite different or is scarce. I was also particularly interested in understanding why models make mistakes. Using explainable AI, I identified cases where models relied on misleading patterns—such as over-associating certain demographic terms with abuse—and designed ways to reduce these biases.

Overall, my work was about making these systems more robust, fair, and adaptable—so they don’t just perform well in theory, but remain reliable in the messy, constantly evolving environments they are meant to serve.

What lessons have you learned from your doctoral experience that you would like to share?

Above all, my PhD taught me resilience and patience. It was a roller coaster—filled with moments of doubt, rejection, and pressure, but also the exhilaration when experiments worked, models improved, or ideas finally clicked. Learning to navigate setbacks, accept constructive criticism, and keep moving forward became just as important as the technical skills I developed.

I was fortunate to have incredible support along the way. My supervisors, Dr. Irina Illina and Dr. Dominique Fohr, and Dr. Nikos Aletras during my research visit to the University of Sheffield, gave me the freedom to explore my ideas independently while providing guidance when needed. Beyond that, my amazing colleagues and friends, especially one post-doc friend — who generously shared their time and wisdom — helped me stay resilient, engage in thoughtful discussions, and keep the bigger picture in mind. These relationships became a crucial support system during the toughest phases of my journey.

One of the most important lessons, though, was learning how to conduct research autonomously: to engage deeply with the state of the art, identify gaps, and develop the intuition needed to generate meaningful insights. That process taught me perseverance, the ability to work through uncertainty, and a “never give up” mindset. By the end of my PhD, that resilience translated into a profound sense of fulfillment—not just in what I had accomplished, but in knowing I had grown as a thinker and a problem solver.

What is your current position and how has your doctorate contributed to that position?

I currently work as an AI Researcher at Vivoka in Metz, a voice AI company developing edge-based platforms that enable developers and organizations—particularly in reliability-critical domains like logistics, manufacturing, and healthcare—to integrate offline voice-controlled interfaces into their systems.

My day-to-day work sits at the intersection of research and real-world constraints. I work on problems in Natural Language Understanding (NLU), Retrieval-Augmented Generation (RAG), and Small Language Models (SLMs) for embedded systems.

What makes this role especially rewarding is seeing research ideas transition into production and have a tangible impact. The environment at Vivoka genuinely values research, giving researchers the space to define directions and take ownership of complex, applied problems.While the PhD was a formal requirement for the role, its deeper value was in the mindset it built. The nature of my work requires a high degree of autonomy—framing problems, exploring possibilities, and operating without clearly defined paths. In industry research, there is rarely a single “correct” answer. Instead, the challenge lies in navigating trade-offs and working with incomplete information. During my doctorate, I developed the intuition to explore uncertain directions and make informed choices among many alternatives. These are the same skills that define how I approach problems today.

When someone asks you for advice on choosing between work experience and Ph.D. studies, what would you say?

It ultimately depends on your long-term goals. A PhD is best suited for those who want to build a career in research—whether in academia or industry—and are motivated by exploring open-ended problems and contributing new knowledge.

Before starting mine, I was working as an AI researcher in the industry and publishing papers, but I felt something was missing: the space to develop my own research intuition and to drive problems independently from start to finish.  That, for me, is what a PhD truly offers. You learn to think critically—often on your own—question existing methods, and explore ideas in depth. Publishing papers is part of the process, but not the main goal.

At the same time, a PhD can be very valuable in industry—especially in roles requiring research, innovation, or solving complex, less well-defined problems. It trains you to handle ambiguity, build long-term vision, and go beyond applying existing tools to actually improving or rethinking them. These are skills that are increasingly important not just in research labs, but also in advanced engineering and product-driven teams.

That said, a PhD isn’t necessary for everyone. If you’re drawn to applied engineering or non-research roles, industry experience may be a better fit. A PhD is a significant commitment, but it can open doors to roles where you shape ideas rather than just implement them—where you identify gaps, think long-term, and take ownership of more exploratory, high-impact work—skills that are especially valuable if you see yourself leading or shaping research in the future.

Any general advice?

A PhD is a journey unlike any other—demanding, often uncertain, but deeply rewarding. My advice to doctoral students is to become your own muse, your own critic, and your own compassion. Remember that breakthroughs often come after breakdowns—moments of struggle and doubt can lay the foundation for your most fulfilling accomplishments and create a story you’ll look back on with pride. The journey is as much about personal growth as it is about scientific achievements, and the skills you develop will stay with you long after the thesis is finished.