Hi, I’m Livia, an MSc graduate in Data Science from Sapienza University of Rome.

My research interests lie at the intersection of Natural Language Processing, Large Language Models (LLMs), Computational Creativity, and Mechanistic Interpretability. I am particularly interested in understanding how language models generate creative text, how humans perceive AI-generated content, and whether creativity-related behaviours can be linked to interpretable internal representations.

My Master's thesis focused on AI-generated Japanese haiku, combining computational creativity, human evaluation, and Japanese NLP to investigate authorship attribution, perceived humanness, and aesthetic judgments of AI-generated poetry. This work naturally led me toward Mechanistic Interpretability, where I hope to better understand how creative behaviours emerge inside language models and how they can eventually be interpreted and controlled.

Alongside my AI research, I also collaborate on projects related to physical activity, health, and behavioural science, contributing to evidence synthesis and research on health-promoting interventions.

• AI Research

  My main research currently focuses on Large Language Models, Mechanistic Interpretability, and Computational Creativity.

  My interests include:

  • Large Language Models
  • Natural Language Processing
  • Mechanistic Interpretability
  • Sparse Autoencoders
  • Activation Steering
  • Representation Learning
  • Human Evaluation of AI-generated text
  • Japanese NLP
  • Controllable Text Generation

• Frontend Development: I’ve been exploring frontend development through beginner-level projects focused on animations and smaller interface designs. You can check out these projects here. While they are not full web applications, they’ve allowed me to practice key skills in HTML, CSS, and JavaScript.

• Digital Illustration and Design: In addition to my technical projects, I’m passionate about digital illustrations and design. I’ve created several design projects and websites, which I showcase on my personal website. You can also see more of my work on my Instagram profile @wakai.design.

• Research

  Alongside my AI research, I am also involved in multidisciplinary research related to physical activity, health, and behavioural science.

  My recent collaborations include:

  • DE-PASS COST Action, contributing to evidence synthesis on modifiable determinants of physical activity and sedentary behaviour in older adults, with the goal of supporting healthier environments and evidence-based policy making.

  • HOPE Project, focusing on physical education and health-related support for children and adolescents in hospital-school settings through systematic evidence synthesis.

• CyberHack4AI: My team and I secured 2nd place at the CyberHack4AI hackathon, held at Sapienza University of Rome in October 2024. Our project, StudentBridge, is an AI-powered platform designed to support international students moving to Rome by providing centralized information on housing, legal documentation, and university processes. You can check out the full project here.

• 🥋 I'm a brown belt in judo, which I’ve practiced as of June 2026 for 11 years. Judo has instilled in me resilience, discipline, and a constant drive for improvement, qualities that I apply to both my academic and professional pursuits.

• 🎹🎻 Music has been a key part of my life. I’ve been playing the piano since I was 8, and I also played the violin during middle school. This blend of technical and creative outlets helps me balance my logical side with artistic expression.

• 🇯🇵 I have a strong interest in Japanese language and culture, which has shaped both my academic journey and my research path toward Japanese NLP, AI-generated haiku, and computational creativity.

• JLPT N4 (Japanese Language Proficiency Test), obtained in 2022
• Cambridge English Level 2 Certificate in ESOL International (Advanced - C1), issued in November 2020
• IELTS (B2), issued in July 2017
• Udemy Certificate In Python for Data Science and Machine Learning: View Certificate

I recently graduated with an MSc in Data Science from Sapienza University of Rome, where I explored machine learning, statistical modelling, Bayesian analysis, and Natural Language Processing.

During the final stage of my Master's degree, Large Language Models became the central focus of my research. Languages have always fascinated me, and I realized that combining linguistics with AI opened fascinating opportunities to study creativity from both computational and human perspectives.

This led to my Master's thesis on AI-generated Japanese haiku, where I investigated how humans perceive AI-generated poetry, authorship attribution, and the factors influencing perceived humanness. More recently, this work has evolved into a growing interest in Mechanistic Interpretability, with the goal of understanding how creativity-related behaviours emerge inside language models and whether they can be interpreted, evaluated, and eventually controlled.

At the same time, I continue collaborating on research in health, physical activity, and behavioural science, applying evidence synthesis methods to projects such as DE-PASS and HOPE, which investigate interventions promoting healthier lifestyles and educational support for children and adolescents in hospital settings.

I also explored the Hikikomori phenomenon, during my Bachelor's thesis. Understanding how social, cultural, and technological factors influence human behaviour remains one of my long-term interests and continues to inspire the interdisciplinary perspective I bring to my research.

My academic path has also been strongly shaped by Japan. During my studies, I had the opportunity to be an exchange research student at Shibaura Institute of Technology in Tokyo, an experience that strengthened both my interest in Japanese NLP and my motivation to continue developing my research career both in Italy and in Japan.

Feel free to reach out if you're interested in collaboration, research, or just to discuss shared interests.

• GitHub statistics : @anuraghazra
  
• Profile counts badge : @antonkomarev