Projects

Ongoing project

NabuPD

AI-Augmented Treatment of Parkinson's Disease Patients

NabuPD is an active clinical pilot being developed with Hadassah Medical Center to address gaps created by infrequent clinic visits and retrospective reporting. It uses a familiar messaging interface and a multi-agent backend to collect and interpret symptom and medication information shared in conversation.

The system is designed to turn routine messages into longitudinal clinical context, route concerns to specialized agents, schedule follow-ups, and give clinicians concise summaries and red alerts. Its design emphasizes identity anonymization before clinical processing, protocol-constrained reasoning, clinician oversight, and guardrails that validate responses and escalate urgent cases. Current research directions include latent-state analysis, longitudinal memory, and multimodal interaction data.

Ongoing project

LoCoPD

A Framework for Longitudinal Conversational Modeling in Parkinson's Disease

LoCoPD is a Parkinson's disease-grounded framework for developing and evaluating conversational agents that follow a patient across multiple sessions. It represents a stable baseline profile and persona alongside a dynamic patient state that can change over days, weeks, and months, allowing generated conversations to reflect reported symptoms as well as partially observed changes.

The project targets three reusable research components: a dynamic longitudinal conversation generator, a curated frozen evaluation dataset, and state-and-memory benchmarks. Together, these are intended to test whether companion agents can track, recall, and reason over evolving patient context, while the frozen dataset lets different systems be evaluated on identical transcripts.

Ongoing project

Tourette Tics

Deciphering the Dynamics of Tics in a Rat Model of Tourette Syndrome

We endeavour to decipher the dynamics of Tourette tics across the cortex and the basal ganglia. By chronically implanting multi-site probes (NeuroPixel) in a rat model of Tourette syndrome, we simultaneously record multiple brain regions—in particular cortex, striatum, and globus pallidus (GP)—in awake, freely moving animals. After inducing tics, we analyse local field potentials (LFP), action potentials (AP), and individual sorted units to infer the circuit dynamics underlying tic events.