I build search and ML systems where the important question is not only whether the model works, but what evidence, memory, or signal it is actually using.

Former Software Engineer at Konan Technology, now completing an MSc in Artificial Intelligence at Queen Mary University of London.

• RAG grounding and retrieval evaluation
• Information retrieval and ranking systems
• Mechanistic analysis of model behaviour
• Production ML / backend engineering

BERT retrieval across six domains with BM25 + FAISS, RRF fusion, ColBERTv2 reranking, and causal context perturbations for grounding evaluation.

Mechanistic analysis of PPO-trained Transformer policies in a partially observable MiniGrid T-maze using retrieval blocking, patching, and source exchange.

Enterprise search and NLP systems for clients including the Supreme Court of Korea and Millie's Library. Supreme Court search latency reduced from 5+ minutes to 8 seconds.

UK BESS price forecasting project selected for QMUL QIncubator. XGBoost reached MAE £4.10/MWh over 55,000+ settlement periods.