My name is Eva, and I am a postdoctoral research associate at the University of Bristol, where I use mathematics, machine learning, and artificial intelligence in financial applications. My academic journey began with a PhD in Computational Finance, where I focused on evolutionary computation for algorithmic trading. In my current position, I have been developing interpretable frameworks for detecting and categorising anomalies in the bond market, while also applying AI in bond market sentiment analysis and exploring approaches to predictive modelling.
My main research interests lie in applying natural language processing to finance, particularly in extracting sentiment and insights from unstructured data to better understand market behaviour, using multi modal datasets. I am motivated by the potential of these methods not only to improve predictive modelling and risk management, but also to contribute to more transparent and trustworthy systems. I am especially interested in the social impact of AI, and how it can support policymakers and society.
Beyond my research, I enjoy working across disciplines, as it can support the creation of impactful methods, connecting technical models with the human element. This ensures that innovation accounts for usability and real world implications. Outside academia, I enjoy reading widely on technology and society, exploring how emerging tools can reshape our understanding and I work on improving my storytelling skills, since I believe effective communication is essential to making research accessible and meaningful.
What are you working on?
I have been working with Prof. John Cartlidge using bond market transaction data from our industry partner Propellant.digital, identifying and categorising reporting errors and market anomalies. Our aim is to build a transparent and explainable framework into understanding such errors and anomalies, and how inaccuracies in transaction data may influence regulatory oversight and market stability. Alongside this, I am exploring how AI and large language models can be applied to bond market sentiment analysis and predictive modelling. These methods have the potential to highlight hidden dynamics in fixed income markets and provide regulators and market participants with tools to better predict risks and opportunities, all while improving transparency and trust in financial systems.
What excites you most about your field of research?
What excites me most is the power of combining computational methods with multimodal data to address real world challenges. Financial markets are complex and influenced by events and human sentiment. Thus, using computational approaches from evolutionary computation to AI allows us to uncover patterns that traditional approaches can miss. I am especially interested in research that has technical complexity and practical relevance, such as identifying hidden risks or improving transparency in systems. Furthermore, I believe the pace of innovation in AI also means there are always new methods to test and adapt, ensuring the area remains interesting, challenging, and impactful.
Who has influenced your academic journey?
My PhD supervisor, Dr. Michael Kampouridis, and my postdoc supervisor, Prof. John Cartlidge, have been hugely influential, especially in how to conduct research, to be detail oriented, and the significance of storytelling. Their guidance showed me the importance of framing research questions in ways that are not just technically sound, but also meaningful for society, regulation, and financial stability. Beyond that, work by researchers in evolutionary computation, machine learning, and AI have influenced how I think about interpretability, robustness, and real world applicability. I particularly admire those who combine quantitative finance with ethical and regulatory concerns, since they do not only consider if something can be done, but how it can be done responsibly.
Can you recommend a paper you think we should read?
I recommend the paper “Your AI, Not Your View: The Bias of LLMs in Investment Analysis” by Lee et al., 2025. This paper introduces a systematic framework for identifying and quantifying confirmation bias in large language models for investment decision making. It demonstrates how models can exhibit hidden preferences, for example favouring large-cap stocks, and how these biases persist even against stronger counter evidence. I believe this paper offers a good perspective and insights into building more transparent and trustworthy financial AI systems.
What are your ‘Top Tips’ for early career researchers?
My main advice to ECRs is to always keep it simple. When a specific part of the problem/project does not necessarily need machine learning or AI models, then start with simple mathematics, and consider increasing complexity from there. While machine learning and AI are powerful, it is easy to fall into the “trap” of adding unnecessary complexity. Furthermore, I believe it is important to collaborate across different disciplines as it opens new perspectives and applications you might otherwise miss. Always aim for clarity and reproducibility, to aim for transparency in methods and data, and to document your code. Reproducibility builds trust and supports impact.
Finally, take care of yourself😊. Balance work and life, accept that setbacks are inevitable, and remember that your curiosity and community will carry you forward.