Projects

Qualitative research methods like semi-structured interviews give clinical researchers rich insight into patient and clinician experience, but analyzing them is slow and labor-intensive. SentinelAI is a model-agnostic, single-agent AI system built to accelerate this process, using Microsoft 365 Copilot to generate a coding framework for each transcript, extract key quotes, and synthesize themes and subthemes into a consolidated table.

We validated SentinelAI against human analysts using interviews from the SENTINEL study, a palliative care intervention for patients undergoing hematopoietic cell transplantation. SentinelAI maintained general consistency with human coding while cutting analysis time by roughly two orders of magnitude, and its highlighted findings aligned with changes later implemented in the study's next phase.

Built on an accessible, non-frontier model, SentinelAI points to a broader opportunity: using LLM agents to close the gap between data collection and clinical implementation across qualitative research.

Complex neural systems are difficult to replicate for experimental purposes. Recent advances in multielectrode arrays, including high-density probes such as Neuropixels, paired with computational spike-sorting algorithms, now allow researchers to isolate single-neuron activity from population-level electrical recordings. These tools give neuroscientists a window into how populations of neurons communicate to produce complex functions, such as speech, and how those circuits are disrupted when pathological cells, such as glioma cells, are present in the surrounding tissue.

Electrophysiological recordings of neuron-glioma co-cultures represent a new frontier in understanding glioma biology, including emerging evidence that glioma cells can functionally integrate with neuronal circuits. This work may ultimately help inform neurosurgeons' decisions around maximal safe resection.