Current Research
I am currently exploring novel modifications of large language model architecture to improve lifelong learning performance, modeled with task/class incremental training, while remaining compatible with current model architectures and training approaches such as QLoRA and MeZO.
Continual Learning in Transformer Models
Designing scalable architectures for lifelong and incremental learning with modern models and fine-tuning techniques.
Context-Aware Multimodal Systems
Designing systems that integrate text, images, and context to reason about subtle human and environmental phenomena.
Recent Publications
CPIQA: Climate Paper Image Question Answering Dataset for Retrieval-Augmented Generation with Context-based Query Expansion
Extracting and Summarizing Evidence of Suicidal Ideation in Social Media Contents Using Large Language Models
Technical Skills
Experience
NLP & Machine Learning Research Intern
Part of the DR-Africa grant worth £358k, we collaborated with the MET Office and the Centre for Machine Intelligence, as well as 2 leading universities in an interdisciplinary project to automate the rescue of climate data from traditional mediums.
- Explored state-of-the-art stable diffusion models for data recovery
- Developed novel prompting paradigm for retrieval-augmented-generation tasks
- Collated multimodal climate science QA dataset for designed for RAG
- Published work as first author to ClimateNLP workshop @ ACL 2025
Laboratory Demonstrator
Delivery of practical teaching to university students from 1st year up to MSc students. Responsible for providing students with feedback while being approachable with any problems.
- Delivered 1-to-1 teaching for students with personalised mentoring
- Explored potential problems prior to teaching, while being ready to handle new issues
Web Development Intern
Responsible for re-designing a dated website, conducting and incorporating user feedback, and implementing new features.
- Built responsive web interface with constrained tech stack
- Implemented new interactive features including PhD project browser
- Identified and conducted user feedback to inform development