Section 3 of 7
Research experience
Trustworthy AI research across LLM evaluation, participatory design, and health applications.
Selected research roles — focused on reliability, evaluation, and high-stakes use.
Research 01
Research Collaborator · Harvard Health System Innovation Lab Nov 2024 – Present
Automated medical research classification at scale.
Outcomes
- Built a few-shot LLM classifier (GPT-4o-mini + RF ensemble) automating classification of a 200,000-record cancer research dataset with 93% human-machine agreement — reducing manual labelling effort by over 80%.
- Designed a precision-first abstention mechanism that eliminated human verification of positive decisions, enabling clinical experts to focus on genuinely ambiguous cases.
- Co-authored: Public and philanthropic research funding, publications, and research networks for cancer in the Commonwealth and globally between 2016 and 2023: a comparative analysis, The Lancet Oncology 26(9), e466–e476, 2025.
- Invited to the Lancet Commissioners meeting to advise on AI's role in global cancer control and research equity.
Research 02
Research Assistant · University of Southampton Jun 2024 – Present
AI agent evaluation, governance, and participatory design.
Focus
- Achieved top ranking at the Concordia Contest 2024 with an LLM agent architecture for cooperative multi-agent interaction — published at NeurIPS Datasets and Benchmarks Track 2025.
- Researching AI governance, explainability, and public engagement in participatory agentic system design — paper under review at ACM CHI.
Research 03
Cooperative AI Summer School 2025
Intensive research training in multi-agent evaluation and cooperation.
Work
- Selected participant; developed evaluation metrics for exploitability in multi-agent systems.
Core themes
◎
Evaluation-first
Define what 'trustworthy' means, then measure it.
⊞
Grounded generation
Retrieval and citations to constrain outputs in high-stakes settings.
◈
Participatory design
Design with the people affected — not just for them.
