PhD Student
I’m a PhD student in Computer Science at The University of Manchester and the National Centre for Text Mining (NaCTeM), supervised by Prof. Sophia Ananiadou. I am partly sponsored by BAE Systems.
I study large language models and agentic systems for software and systems engineering, with a broader interest in reliable AI in high-risk domains. My work also explores financial NLP and multilingual evaluation.
New preprint: Janus studies how goals can distort the way LLMs communicate factual information.
I moderated the AI Power Hour roundtable at The Millennium Alliance’s C-suite assembly in Madrid, discussing value, trust, and hidden complexity of AI.
Two papers accepted to ACL 2026: MultiFinBen (Main conference) and Same Claim, Different Judgment (Findings).
One paper accepted to LREC 2026: Semantic Label Drift in Cross-Cultural Translation.
Preprint · 2026
LLMs can mislead without technically lying. We introduce Janus, a benchmark for measuring how models reshape the same underlying evidence to create misleading impressions when pursuing an institutional goal.
Preprint · 2026
A trading strategy can fail because of what it chooses to trade or how it executes. We use language-based feedback to help LLMs improve these two decisions separately.
ACL · 2026
Financial information crosses languages, tables, images, and audio. We build MultiFinBen to test whether LLMs can handle that variety, including the harder tasks that simpler benchmarks miss.
Findings of ACL · 2026
The same financial claim should receive a consistent judgment. We build a multilingual benchmark to test how a model’s assigned role, personality, or social context changes its verdict.
MODELSWARD · 2026
Turning written requirements into a UML diagram involves several distinct decisions. We introduce NOMAD, a multi-agent framework that distributes these decisions across specialized LLM agents to improve modeling performance and systematically characterize remaining errors.
LREC · 2026
A translated text can acquire a different sentiment label. We study how cultural differences drive this shift and why an LLM’s cultural knowledge can sometimes amplify it.
Natural Language Processing Journal · 2025
Emotions often occur together, yet classifiers typically predict each one independently. We introduce emotion anchors that help each classifier account for these relationships, including few-shot domain adaptation.
EMNLP · 2025
Greek financial text poses challenges that multilingual models often miss. We introduce a Greek financial benchmark, expert-annotated datasets, and a specialised language model to study and address this gap.
DeTermIt! at LREC-COLING · 2024
Clinical-trial results are difficult to navigate and understand. We create CARES and study how to turn long reports into plain-language summaries, examining the factual errors models make along the way.
— Present
PhD in Computer Science
Language models, multimodal NLP, and agentic systems. Supervised by Prof. Sophia Ananiadou; sponsored by BAE Systems.
—
Researcher
Emotion detection and sentiment analysis with Prof. Ioannis Pitas. Led integration of the sentiment-analysis module for the Horizon Europe TEMA project.
—
M.Eng. in Electrical & Computer Engineering
Five-year integrated master’s. GPA: 9.00/10.00 (top 1%). Diploma thesis on plain-language summarisation of clinical trials, supervised by Prof. Grigorios Tsoumakas.
Get in touch
For collaborations, research ideas, or just an e-coffee.
<firstname> dot <lastname> at manchester.ac.uk
