Skip to content
FabioAlmeida
Back to projects

AI-Powered Applications

GlucoAlarm

Role: Solo architect and developer — product design, backend, integrations, and mobile app.

1. Challenge

Families managing a chronic health condition that requires continuous glucose monitoring need real-time alerts when readings cross dangerous thresholds — existing monitoring apps notify only the primary phone, with no redundancy if that alert is missed.

2. Business Requirements

  • Sub-minute alert delivery with redundant channels (WhatsApp, SMS, and voice call escalation) so a critical alert is never silently missed.
  • Tiered severity logic distinguishing routine, elevated, and emergency glucose events.
  • AI-generated insight summaries that support awareness without ever issuing dosing or clinical recommendations.
  • Multi-tenant account model with subscription billing and doctor/caregiver read-only access.

3. Proposed Architecture

An edge-deployed ingestion service pulls continuous monitor readings on a short polling interval, evaluates them against per-user threshold rules, and routes qualifying events through a tiered notification pipeline (WhatsApp first, SMS and automated voice calls as escalation fallbacks) backed by a serverless database for account, device, and alert-history state.

4. Implementation Approach

Shipped iteratively: core ingestion and threshold alerting first, then multi-channel escalation, then AI-generated trend summaries with explicit safety guardrails, then a native iOS app and multi-tenant billing to convert the tool into a subscription product.

5. Security Considerations

Treats all health data as sensitive by default: access is scoped per account, caregiver/doctor access is read-only, and the AI insight layer is constrained from producing medical dosing guidance under any prompt path.

6. Results

Operates as a live, real-world safety tool with redundant alert delivery, tiered severity, and production incident response experience (including hardening delivery pipelines after real-world load and quota issues). Converted from a personal tool into a subscription SaaS product.

7. Technologies Used

  • Next.js
  • TypeScript
  • Cloudflare Workers
  • D1
  • Twilio
  • WhatsApp Business API
  • Anthropic Claude
  • iOS (TestFlight)