At Macquarie University, I independently designed and ran experiments on computer vision, LLMs, and precision agriculture. I benchmarked frontier vision-language models, and published the results in The Knowledge Engineering Review.
I built a pipeline to evaluate multimodal LLMs on subspecies-level identification of crop cultivars and pests. I analysed failure modes, including unreliable structured output and overconfident predictions, and added conformal prediction to calibrate model confidence.
I fine-tuned a language model to assist with automated software testing, and also fine-tuned a model for short-fiction generation. At Singapore Management University, I conducted research on large language models and transformer architectures, exploring instruction tuning and DPO for domain-specific alignment.
I built the backend for an Android AI scam-call detection app, handling call transcripts, user sessions, and incoming-data validation. I also built its rule-based policy component, evidence log, performance metrics, and Docker-based deployment.

