Care Call AI

Care seen. Needs heard. Help delivered.

• Live application: care.alexraixon.com
• Project landing: about.care.alexraixon.com
• Demo video: Watch on YouTube
• Public source: github.com/NeoSPU/care-call-ai
• CI results: CareCall CI
• CALL-E contribution PR: CALLE-AI/awesome-phone-call-agents#550
• CALL-E contribution package: contribution/awesome-phone-call-agents/

The live application is deployed from the private production repository. This repository is the synchronized, sanitized, runnable hackathon edition used for technical review.

Care Call AI is a practical support outreach and fulfilment coordination workflow built with CALL-E. It helps charities, disability support organizations, and community support services run approved phone check-ins with beneficiaries and turn completed conversations into practical support requests and printable fulfilment orders.

The product is built around a simple belief: a care call only matters if the person's request is heard accurately and becomes an action the team can complete.

• Care seen - dashboard statistics show recipient readiness, safety categories, condition mix, service demand, urgent callback pressure, and cases that need human attention.
• Needs heard - CALL-E check-ins are prepared from recipient context and only started after no-call preflight plus explicit coordinator approval.
• Help delivered - completed calls become structured requests, review rows, order history, and print-ready handoff sheets.

Charities and support organizations spend a lot of time contacting beneficiaries, listening carefully, writing down everyday requests, and routing them to food and groceries, medication pickup, transport, cleaning, laundry, other home help, companionship, leisure activities, repairs, documents support, or other non-clinical support teams.

The risk is not only that a call is missed. The larger risk is that a person is heard in the moment, but their request is later delayed, duplicated, mistranscribed, routed to the wrong team, or forgotten.

Care Call AI uses CALL-E inside a safer coordinator-led workflow:

• recipient safety categories;
• condition-aware call goals;
• trusted answerers;
• no-call preflight;
• exact approval before live calls;
• repeat-call and daily-limit safeguards;
• structured service request generation;
• review states for uncertain, restricted, or no-request outcomes.

Care Call AI lets coordinators:

• review a daily care population before outreach starts;
• identify eligible, blocked, critical, and operator-only recipients;
• inspect and update recipient care cards;
• prepare a CALL-E outreach batch from approved recipient context;
• run preflight without placing calls;
• approve a real CALL-E batch only after explicit confirmation;
• import completed CALL-E results with transcript evidence and summaries;
• extract explicit practical needs without inventing unsupported requests;
• keep review-only and no-request outcomes visible;
• print grouped fulfilment sheets with checkboxes and signoff lines;
• monitor urgent callback requests in a separate queue;
• accept token-protected Siri Shortcut callback requests for registered recipients;

• /dashboard - Care seen statistics.
• /dashboard/operator - Needs heard auto-call round preparation.
• /dashboard/preflight - no-call preflight and approval gate.
• /dashboard/recipients - recipient cards.
• /dashboard/urgent-callback - priority callback queue.
• /dashboard/orders/print - Help delivered summary and printable orders.
• /support, /privacy, /terms - public support and safety pages.

The project is safe by default:

• tests use fake CALL-E runners;
• no-call preflight places zero calls;
• live calls are disabled unless explicitly enabled;
• maximum live batch size is 1;
• recipient-triggered callbacks have daily limits;
• repeated recent calls are blocked from unattended automation;
• real phones are not committed;
• protected backend calls require a bearer token;
• browser code does not receive backend, CALL-E, or support delivery credentials.

Care Call AI is not a medical, healthcare, clinical, patient-care, or emergency- response product, and it is not intended for hospitals, medical institutions, or emergency services. It does not replace clinicians, carers, support workers, or human coordinators.

The judging setup requires only Git, Docker Desktop, and Make. Clone the public repository and enter it:

git clone https://github.com/NeoSPU/care-call-ai.git
cd care-call-ai

Create the local demo configuration:

cp .env.example .env.local

The supplied file already contains every value required for local review:

CARECALL_OPERATOR_USERNAME=carecall-coordinator
CARECALL_OPERATOR_PASSWORD=carecall-demo-password
CARECALL_AUTH_SECRET=carecall-local-development-secret-not-for-production
CARECALL_BACKEND_API_TOKEN=carecall-local-backend-token
CARECALL_LIVE_CALLS_ENABLED=true
CARECALL_CALLE_API_KEY=
CARECALL_SIRI_CALLBACK_TOKENS=

The fixed values are intentionally public and local-only. They are not production credentials. The empty CALL-E and Siri values are secret-placement slots: never commit real values.

The application starts and can be reviewed with CARECALL_CALLE_API_KEY left empty. In that state, the backend fails closed and cannot place an outbound call. To verify the complete real-call path, put your own official CALL-E dashboard Access Key in .env.local:

CARECALL_CALLE_API_KEY=<your-official-CALL-E-access-key>

CARECALL_LIVE_CALLS_ENABLED=true, the API provider, API URL, and the one-call batch limit are already configured for the local Docker stack. After changing the key, restart with make demo-down followed by make demo-up. A real call still starts only after the coordinator selects an eligible recipient, reviews the no-call preflight, completes all confirmations, enters EXECUTE LIVE CALLS, and presses Start calls now. Use only a consented recipient and a phone number you are authorized to call.

Build and start the complete frontend and backend stack:

make demo-up

Verify the local backend:

make demo-smoke

Open http://localhost:3000 and sign in with:

Username: carecall-coordinator
Password: carecall-demo-password

Stop the demo when finished:

make demo-down

The public demo uses a deliberately simple shape:

Next.js frontend
  -> same-origin server routes
  -> protected Python backend
  -> database-backed recipient, call, callback, and order records
  -> guarded CALL-E execution
  -> conservative call-result import
  -> Help delivered orders and printable sheets

This repository is a sanitized public hackathon edition. It keeps the working CALL-E workflow and removes private deployment scripts, internal planning documents, real participant data, and production-specific infrastructure notes.

Run the product gate:

make test

Useful individual checks:

make backend-test
make frontend-test
make frontend-build
make secrets-check

Siri through Apple Shortcuts is the verified voice-assistant callback integration in the current version. A registered and consented recipient can say Hey Siri, Raixon Callback to request a protected CALL-E callback. The request still passes CareCall's eligibility, daily-limit, run-tracking, result-import, and human-review controls.

Generate a different token for each registered recipient:

openssl rand -base64 48

For an optional local callback test, replace the empty token value in .env.local with a recipient mapping, then restart the local demo:

CARECALL_SIRI_CALLBACK_TOKENS=rec-001=<recipient-callback-token>

This token is not needed for the normal local judging path. Never commit it or show it in demo screenshots.

1. Open Apple Shortcuts on the recipient's iPhone or iPad, tap +, and name the Shortcut Raixon Callback.

2. Add Get Contents of URL and set the URL to:

   http://<computer-LAN-IP>:3000/api/callback-requests
   

3. Expand the action details and select POST with a JSON request body.

4. Add these headers:

   Authorization: Bearer <recipient-callback-token>
   Content-Type: application/json
   

5. Add one JSON row per field (do not paste the whole object into one key):

   Key: request_text  Type: Text  Value: Please call me back
   Key: locale        Type: Text  Value: en-GB
   Key: device_label  Type: Text  Value: Recipient iPhone
   

   Omit recipient_id; the server derives it from the bearer token.

6. Add Get Dictionary Value for the response field message, followed by Speak Text or Show Result.

7. Run the Shortcut once inside Shortcuts and confirm that CareCall accepts the request. Then test the phrase Hey Siri, Raixon Callback.

After frontend token validation, the protected backend starts an immediate CALL-E callback for eligible recipients and records the linked run in the Urgent Callback queue. The MVP default is no more than three automatic recipient-triggered callbacks per recipient per day. It is not an emergency medical service.

The Siri Shortcut path is another way to enter the same safety model, not a way to bypass it.

Siri through Apple Shortcuts is the verified callback integration in the current version. The platform-specific trigger is separated from the protected callback endpoint and CALL-E workflow, so planned adapters for Alexa, Google Assistant, Yandex Alice, and other voice assistants can reuse the same consent, eligibility, callback-limit, audit, and human-review controls. Those additional voice-assistant integrations are not currently available.

Siri and Apple Shortcuts are trademarks of Apple Inc. CareCall AI is not affiliated with or endorsed by Apple.

CALL-E contribution material:

contribution/awesome-phone-call-agents/

Suggested contribution areas:

• Apps for the full Care Call AI operator workflow.
• Agent Skills for the reusable CareCall intake skill in agent-skills/carecall-intake/SKILL.md.

Care Call AI is a practical-support outreach and service-request workflow that helps coordinators use CALL-E responsibly for approved outreach. Critical, blocked, do-not-call, repeated too recently, and operator-only recipients are excluded from unattended automatic calling. Unsupported, illegal, unsafe, exploitative, age-restricted, or region-restricted requests must not become fulfilment orders.