I build and evaluate AI systems for legal text at Durham University — retrieval and reranking for statutes and case law, retrieval-augmented prompting, and rigorous evaluation of large language models for legal reasoning.
Team DU participated in all five COLIEE 2026 tasks. Task 4 was our strongest official result, ranking first overall in statute entailment.
Introduces UK-STATUTECORPUS and Distilled-Voyage-ModernBERT for provision-level statutory retrieval.
A unified empirical study of Team DU systems across legal case retrieval, case entailment, statute retrieval and entailment, and legal judgment prediction.
Ranked first of ten teams and awarded first prize. The paper also shows that re-tuning decision thresholds on validation data makes held-out results worse.
Shows that the correct tool was always listed first, so a model could score well by position alone. Shuffling the options cut accuracy by 20 points, and the organisers’ overview reports the same shortcut.
My work sits at the intersection of artificial intelligence, natural language processing, and legal information systems. I study how models retrieve legal materials, rank evidence, and make entailment decisions under the constraints of legal language.
Team DU · official & post-competition results
Completed one peer review for Expert Systems with Applications, an Elsevier journal on expert and intelligent systems applied worldwide.
View certificate →
Completed a peer review for the Journal of Computational Law and Legal Technology, a journal on computational methods and AI for law, acknowledged by the Editor-in-Chief.
View acknowledgement →
Presented “Enhancing Trust in Legal AI: Optimising Span-Level Retrieval Architectures on LegalBench-RAG.”
Happy to connect with researchers and collaborators interested in Legal AI, NLP, information retrieval, reranking, and trustworthy AI systems for legal text.
