CareBuddy

The voice-first AI companion that closes the chronic-care loop.

CareBuddy helps people managing multiple long-term conditions turn fragmented home-health information into one safe next action. It verifies what changed, interprets it across the patient’s conditions and care plan, follows up to see whether the action worked, and brings in a human care team only when necessary.

Click to watch the CareBuddy demo.

Hackathon prototype

The seeded demo follows Sara, 58, who manages type 2 diabetes, hypertension, and early chronic kidney disease.

Sara reports dizziness, an elevated blood-pressure reading, and a late medication by voice. CareBuddy confirms the extracted information, checks measurement quality and multimorbidity context, gives one governed next action, and schedules a repeat. When the repeat remains unresolved, the app prepares a concise clinician brief. A clinician approves a message, Sara receives the next step, and CareBuddy continues following the outcome.

The core loop is:

OBSERVE → VERIFY → INTERPRET → ACT → FOLLOW UP

Why it is different

• Closes the loop: advice becomes a tracked action with an outcome.
• Voice first: patients can speak naturally instead of completing long forms.
• Multimorbidity aware: context is shared across diabetes, hypertension, CKD, symptoms, medications, and personal targets.
• One next action: the patient does not need to interpret a clinical dashboard.
• Graduated human involvement: clinicians see unresolved exceptions, not every stable reading.
• Governed AI: deterministic rules own safety and escalation; AI extracts and explains.
• Demo reliable: every important state is fully seeded and has an offline fallback.

Demo roles

• Sara — Patient: Today’s Focus, voice check-in, confirmation, repeat task, clinician message.
• Maya — Family: reassurance without raw clinical detail.
• Dr Patel — Care team: unresolved worklist, evidence brief, approve/send, resolution.

Architecture

CareBuddy is a native iOS application built with:

• SwiftUI for the patient, family, and clinician experiences;
• SwiftData for profiles, observations, care loops, and review events;
• HealthKit-compatible abstractions with fully seeded demo data;
• a deterministic verification and escalation engine;
• OpenRouter for structured extraction and plain-language explanations;
• cached, schema-validated, deterministic fallback outputs;
• CareBuddy’s existing trends, AI service, NHS content, and PDF foundations.

The language model does not diagnose, prescribe, change medication, or set the escalation level.

Project structure

CareBuddy/
├── App/                    # Entry point, dependencies, SwiftData
├── Core/                   # Models, navigation, theme, resources
├── Features/               # AI, dashboard, trends, brief, onboarding
├── Services/               # HealthKit and assessment orchestration
└── Config/                 # App configuration and secret template

CareBuddyTests/           # Existing unit-test foundation
docs/                       # Pitch, slide prompt, and archived source plan
images/                     # Pitch and README visuals

Running the copied foundation

1. Open CareBuddy.xcodeproj in Xcode.
2. Select an iPhone simulator.
3. Copy CareBuddy/Config/Secrets.example.xcconfig to Secrets.xcconfig only when configuring a private local key.
4. Keep secrets out of Git.
5. Build and run the app.

Demo reliability rules

• Start from a seeded Sara profile.
• Use one-tap stable, check, unresolved, approved, urgent, and offline scenarios.
• Reset scenario state before every pitch.
• Never depend on live HealthKit or network access.
• Confirm voice-extracted data before saving it.
• Validate every model response against a strict local schema.
• Use deterministic fallback copy when OpenRouter is unavailable.

Product boundary

CareBuddy supports organization, self-management tasks, trend interpretation, and clinician-reviewed communication. It does not independently diagnose, prescribe treatment, alter medication doses, control medical devices, or claim continuous clinical monitoring.