Research
Every project leads with a burning question. The code is downstream.
MindTrace AI
An agentic SwiftUI tool that intercepts moments of cognitive offloading and replaces them with calibrated friction loops. We study whether forcing the user to commit to a hypothesis before the model answers improves long-term retention and metacognition. Early study (n=42) shows a 31% lift in 7-day recall vs. a vanilla chat baseline.
Conceptual erosion under LLM offloading
A controlled study probing whether sustained reliance on conversational LLMs erodes users' ability to re-derive concepts unaided over a 14-day window. Pre/post diagnostics across a matched cohort isolate the delta between augmented and unaugmented recall — quantifying the cost of frictionless answers.
Academic endorsements & target venue
Methodology endorsed by leading AI-in-education researchers — Prof. Rose Luckin (UCL Knowledge Lab) and Prof. Neil Heffernan (WPI · ASSISTments) — who reviewed the conceptual-erosion protocol and its instrumentation. The work is currently positioned for submission to AI in Education (AIED) venues.
Neurova
A research scaffold that fuses passive sensor signals (typing cadence, focus-window dwell, ambient sound) into a real-time attention budget. Agents can then schedule interruptions only when the marginal cost is low. Built on a Vertex AI pipeline; a SwiftUI menu-bar app exposes the live signal.