CareRoute Guardian for Alexa+

Evidence-backed care-journey orchestration that verifies every handoff from home to recurring care and back.

Comment

Inspiration

Recurring medical transportation has a dangerous coordination problem: systems can know that an action was attempted without knowing whether the real-world handoff actually happened. A ride can be dispatched without arriving. A booking can be confirmed without pickup. A return ride can fail after an appointment is over.

CareRoute Guardian explores a simple rule: verify before you advance the journey.

What it does

CareRoute Guardian is a stateful Alexa+ simulated experience for a dialysis patient traveling from home to an 8:00 AM appointment and back.

HOME → TRANSPORTATION → CLINIC → APPOINTMENT HANDOFF → RETURN TRANSPORTATION → HOME

When the scheduled 7:15 AM ride does not arrive, CareRoute does not optimistically mark transportation complete. It records the attempted check, captures evidence that pickup was not confirmed, blocks downstream travel states, informs the authorized caregiver, and offers the approved backup workflow.

When a replacement ride is confirmed, pickup remains pending until separate evidence arrives. Clinic arrival is verified independently. Later, when the return ride is dispatched but pickup is not confirmed, the system fails closed again: home arrival stays blocked and a predefined escalation policy routes the unresolved exception to human review.

How we built it

We used the competition-permitted Alexa+ simulated web-experience path. The judge-facing Alexa+ interaction layer is simulated, while the underlying orchestration is working deterministic code:

  • static HTML/CSS judge interface
  • JavaScript ES-module state/reconciliation engine
  • browser localStorage for durable journey state across reloads
  • synthetic transport, clinic, caregiver, and wearable/context adapters
  • deterministic evidence ledger and auditable event history
  • explicit action prerequisites and fail-closed transitions
  • Node built-in automated test suite with zero runtime dependencies

The core rule is:

VERIFY → RECONCILE → ESCALATE OR CONTINUE

Statuses are explicit: PENDING, ATTEMPTED, FAILED, CONFIRMED, BLOCKED, NEEDS_REVIEW, and COMPLETED.

What makes it agentic

This is not a single-turn Q&A bot. Alexa+ is the conversational control surface for a multi-step workflow that maintains state across a care journey, reconciles external outcomes against prerequisites, preserves unresolved work, and explains why the next action is or is not allowed.

Safety and privacy boundary

This project uses synthetic data only. It does not diagnose illness, determine clinical treatment, detect medical emergencies, determine medical safety, or replace healthcare professionals.

The wearable/context panel demonstrates user-authorized wellness/context signals without clinical interpretation. Escalations are operational and policy-based, such as a missing transportation confirmation, and may require human review.

No production medical records, real patient data, or real emergency dispatches are used.

Challenges we ran into

The most important implementation challenge was understanding the Alexa+ simulation path. The organizer materials clarify that hackathon participants may build a simulated Alexa+ web experience without gated preview tooling. We therefore concentrated engineering effort on the real orchestration behavior behind the interaction: state, evidence, reconciliation, safety gates, persistence, and explainability.

Accomplishments we are proud of

  • Booking, pickup, clinic arrival, return pickup, and home arrival are not collapsed into one optimistic status.
  • Every protected transition is evidence-backed.
  • The return-trip exception visibly demonstrates fail-closed behavior and human review.
  • The build has zero runtime dependencies and a 24-test deterministic suite.
  • The architecture is provider-agnostic without claiming affiliation with any transport or wearable company.

What we learned

Agentic systems become more trustworthy when they model uncertainty explicitly. Attempted and confirmed are not synonyms. Durable state and evidence lineage matter as much as conversational fluency when an assistant coordinates real-world tasks.

What's next

A production roadmap would add authorized provider integrations, secure identity and consent, clinic/caregiver interfaces, policy configuration, audit controls, privacy/security review, and domain-specific validation. The same verified-handoff architecture could later support chemotherapy, physical therapy, senior transportation, post-operative mobility, disability transportation, chronic-care visits, and home-health transitions.

Built With

Share this project:

Updates

Submission history