I am a Research Associate in the Data, Uncertainty, Constraints and Knowledge Lab at Imperial College London working on constraining Large Language Models (LLMs) for theorem proving. Recently, I completed my PhD studies in Computer Science at the University of Oxford, under the supervision of Professor Thomas Lukasiewicz. In my research, I develop neuro-symbolic methods that integrate background knowledge constraints into neural networks, enforcing them during both training and inference.
During my PhD, I demonstrated that complex constraints—from linear inequalities to disjunctions over linear inequalities that model non-convex and even disconnected spaces—can be successfully integrated during training to enhance the quality of synthetic data. Building on this foundation, my current work focuses on building similar constraint-enforcement frameworks that can be integrated into LLMs to enhance their reasoning and theorem-proving capabilities. My broader research vision is to bridge the gap between neuro-symbolic AI and real-world applications to build more robust and trustworthy systems.
Previously, I was part of the research team at Five AI, where I worked on detecting reflective symmetries in 3D models. I completed my Bachelor’s and Master’s at The University of Edinburgh, where my Master’s thesis, supervised by Professor Sharon Goldwater, focused on speech-to-text machine translation.
PiShield
PiShield — the first PyTorch package for embedding requirements directly into neural network topologies, guaranteeing that model outputs satisfy the given constraints at both training and inference time. It supports linear inequality constraints, quantifier-free linear real arithmetic, and propositional-logic constraints, with demonstrated gains across autonomous driving, tabular data generation, and functional genomics.
I created PiShield in 2024 during my DPhil at the University of Oxford, and I have been its sole maintainer ever since. Try it out in Colab, or install it locally via pip install pishield.
News
- May 2026. Presented my work at the OxBridge Women in CS Conference (selected for oral presentation).
- May 2026. Our paper Can I Have Your Order? Monte-Carlo Tree Search for Slot Filling Ordering in Diffusion Language Models has been accepted to ICML 2026.
- February 2026. Our paper A Survey on Deep Learning Approaches for Tabular Data Generation: Utility, Alignment, Fidelity, Privacy, Diversity, and Beyond has been accepted to the Transactions on Machine Learning Research (TMLR).
- December 2025. Successfully defended my DPhil thesis in Computer Science.
- October 2025. Started a postdoctoral position at Imperial College London.
- October 2025. Submitted my DPhil thesis in Computer Science at the University of Oxford.
- May 2025. Received the Oxford PhD Runner-up Prize awarded by G-Research.
- April 2025. Presented at ICLR my recent work on deep generative modelling with constraints captured as quantifier-free linear real arithmetic formulae.
- February 2025. Attended the Dagstuhl Seminar on Logic and Neural Networks and presented the work I conducted during my PhD.