My work spans two areas: multi-agent path finding — routing fleets through shared space — and applied large language models in healthcare. A short note accompanies each paper; follow the links for the full text.
A first-authored pilot study comparing three LLMs (GPT-4o, Llama-3.1, Gemma-2) against three traditional machine-translation tools for translating English medical summaries into Arabic, Chinese, and Vietnamese. Traditional MT led on surface metrics, while LLMs showed relative strengths in semantic similarity — underscoring the need for human review before clinical use.
C-ORCA* and C-ORCA*-MAPF are continuous-space multi-agent path-finding algorithms that prevent deadlocks by reasoning over agents' full trajectories, removing the post-hoc correction overhead of prior real-time solvers. They cut expensive MAPF fallbacks by 3–4 orders of magnitude and sustain a ~100% success rate at 200 agents — at near-zero per-timestep overhead.
A peer-reviewed pilot evaluating GPT-3.5, GPT-4, and LLaMA 2 for zero-shot summarisation of doctor–patient palliative-care consultations. Summary quality was benchmarked with BLEU, ROUGE-L, METEOR, and BERTScore alongside readability scoring — GPT-4 led on most metrics, while LLaMA 70B produced the simplest, most readable summaries for patients.
I'm the lead organiser and main contact for theWorkshop on Multi-Agent Path Finding at AAAI-26, where I coordinated a four-person team and the review of 40+ submissions. For the full record of teaching, supervision, and project work, see my CV.