AI Safety · AI Governance · Agentic AI · Foundation Models · NLP
I build trustworthy AI for healthcare and science: conversational agents, clinical NLP, evaluation & safety, and foundation models — with interests in AI safety, governance, and large‑scale data analytics.
Multilingual mental-health conversational agents built with agentic AI, in collaboration with Freedom from Torture. Designing trauma-informed, culturally sensitive systems with evidence-based therapeutic strategies and safe, interpretable reasoning workflows.
LangGraph-driven multi-agent system automating literature reviews across arXiv and PubMed, with metadata enrichment from Crossref. Integrated agents (SearchAgent, SummarizerAgent, CriticAgent, ComparatorAgent) enabling structured synthesis, gap analysis, and future work identification. Supports multiple output formats (Markdown, JSON, CSV) with CLI and REST API access, plus a React/Vite/Tailwind frontend.
Improved BERT for Aspect Extraction, Sentiment Classification, and Review Comprehension. Combined adversarial training with P-SUM/H-SUM, surpassing prior state-of-the-art in domain-specific NLP. Demonstrated efficiency of fine-tuned BERT over LLMs for targeted, low-resource tasks.
Implemented zero-shot pipelines for single-cell RNA-seq analysis, enabling malignant vs. normal classification and multi-class cell state prediction across ovarian, prostate, kidney, and pancreas cancer datasets. Built preprocessing workflows and integrated evaluation with Accuracy, F1, AUROC, and AUPRC metrics. Developed visualisation tools (UMAP, performance plots) to assess model generalisation and batch effects.
Built an AI-powered counselling platform to support students applying abroad (UK, US, Australia). Implemented essay feedback using GPT-4, application timelines, portfolio builder, and AI chatbot for personalised guidance. Full-stack development with React (TypeScript), FastAPI (Python), PostgreSQL, Firebase, and OpenAI API.
Developed an AI-driven emotional support app featuring a built-in chatbot to help users explore their thoughts, with breathing exercises, a crisis helpline, and customisable themes. Currently in development; this open-source project aims to provide accessible mental health support.
Authored a book on Python Programming, distributed over 30,000 copies to uplift underprivileged students.
Outside of research, I invest in people, endurance, and purpose.
I love to teach. Whether through formal TA roles, mentoring peers, or writing accessible resources, sharing knowledge is one of the most rewarding things I do. Seeing someone grasp a difficult concept — that never gets old.
I do long-distance cycling — covering hundreds of kilometres at a stretch. There's something uniquely clarifying about extended time on the road; it's where I do some of my best thinking.
I compete in duathlons and take on charity challenges — combining a love of multisport endurance events with the chance to raise funds and awareness for causes I care about.