2025–present
Assistant to the Caregiver
An early-stage Android application testing whether low-friction capture and caregiver-reviewed AI can reduce the bookkeeping burden between medical appointments.
- Status
- Working prototype
- Evidence
- Formative and limited
- Claims checked
The project doubles as a working test bed for the evaluation questions I help other teams with. Building something that has to earn a caregiver’s trust is a more demanding exercise than reviewing someone else’s system, and the constraints show up faster.
The situation
Family caregivers carry an administrative load that nobody designed and nobody owns. Between appointments they are expected to track symptoms, medications, instructions, and questions, then reconstruct all of it under time pressure in a fifteen-minute visit. The burden is real, it is poorly supported by existing tools, and it falls hardest on people who are already stretched.
My role
I am designing and building the application, and I own the research and evaluation questions behind it: what caregivers actually need, what the system should refuse to do, and what evidence would show that it helps rather than merely appears helpful.
The initial assumption
The project began from a capture premise — that the core difficulty was recording information between appointments, and that making capture fast enough would reduce the burden.
What the evidence changed
Early caregiver conversations moved the emphasis from capture to review and trust. What mattered was less whether something could be recorded than whether a caregiver could see where a piece of information came from, correct it when it was wrong, and decide whether to rely on it at all. That reframing is why the design centres on provenance, uncertainty, correction, and explicit human confirmation for anything consequential, rather than on automation.
Record / 1
What was produced
- A working Android prototype
- A design centred on caregiver review, provenance, uncertainty, correction, and human confirmation for consequential information
- Onboarding and UI/UX revisions following the July 2026 competition application
- A personalized education-card experience, working but still being refined
- Cross-phone synchronization
- Keystore and application distribution infrastructure
Record / 2
Evidence available
- A working prototype that can be demonstrated directly
- Notes and design rationale from early caregiver conversations and formative walkthroughs
- A development history showing the sequence of decisions and what prompted each revision
Credibility requires a boundary
Limits
Caregiver input to date has been formative and limited. It is not a representative validation study, and nothing here demonstrates a reduction in caregiver burden. Controlled walkthroughs are not evidence of reliable long-term use in homes or clinical settings. Synchronization and distribution work are development milestones, not proof of production-grade security, privacy, recovery, or reliability. The application has no clinical validation or endorsement, does not diagnose or recommend treatment, and I make no claim of regulatory, security, privacy, or accessibility compliance. Market, business model, and production-readiness questions remain open, and the feature set moves — ask me for the current state rather than relying on this page.
Transferable value
This is the discipline I bring to an AI product that has to be trusted: deciding what the system must not do, designing where a person stays in control, making uncertainty and provenance visible instead of hiding them behind a confident answer, and defining in advance what evidence would count as the thing working.