I build AI-native research tools at the edge of biomedical engineering, human-computer interaction, and local-first software.

My work often starts from messy, real-world signals: eye movements, vestibular tests, medical images, clinical reports, coding sessions, study notes, and personal workflows. I like turning those materials into systems that can be inspected, reused, and improved — not just demos that work once.

Across my projects, I move between model experiments, data pipelines, interface design, deployment constraints, and documentation. I have built tools for gaze estimation, nystagmus monitoring, vestibular report generation, radiomics workflows, AI coding analytics, personal knowledge systems, and small desktop/mobile products.

What connects these projects is not a single stack. It is a way of working: make the context visible, reduce hidden state, keep the system runnable, and let human judgment stay close to the machine output.

I am especially interested in biomedical AI and agentic software that remain grounded in real data, real users, and real constraints. For me, good AI systems are not just accurate models; they are workflows people can trust, question, adapt, and carry forward.