I'm an AI engineer shipping production LLM systems. That means end-to-end agentic workflows with custom MCP servers, multi-step tool orchestration, and RAG pipelines grounded in verified entities to reduce hallucination. Recent work includes a neuro-symbolic agent for spoken Tamil morphology that grounds LLM reasoning in a Neo4j knowledge graph, and a clinical entity-linking RAG pipeline that outperformed ClinicalBERT and direct LLM generation by constraining outputs to real ontology candidates.

I came to engineering through research. My PhD in applied mathematics (NLP focus) from the University of Adelaide produced peer-reviewed publications in top journals and conferences on patient-reported experiences in healthcare, including a probabilistic emotion-recommender system evaluated with IR-style metrics. That background shapes how I build now. I default to rigorous evaluation, think carefully about calibration and uncertainty, and reach for grounded retrieval over direct generation when faithfulness matters.

What I build for fun: Tamil learning tools (my wife and son are Tamil, so partly to help raise him bilingual, partly to understand what my wife is saying to him), Australian wildlife photography (a particular weakness for small birds), native gardening, and conservation.