Inspiration

Rural and peri-urban water infrastructure teams often operate in low-connectivity environments where borehole and water-point failures go unrecorded or lack structured triage data. MajiPulse enables field reporters to log incidents offline while giving maintenance technicians and administrators AI-assisted triage and cross-incident pattern intelligence.

What it does

  • Offline-First Incident Reporting: Field reporters capture water-point failures offline in an organization- and user-scoped IndexedDB outbox with deterministic, idempotent Firestore synchronization and conflict resolution.
  • Privacy-First Hybrid AI Extraction: Before any model sees free-text descriptions, a deterministic redactor strips phone numbers, emails, IDs, and personal identifiers and displays a live preview to the reporter. MajiPulse then uses Firebase AI Logic (PREFER_ON_DEVICE) to extract structured incident fields locally via Gemini Nano in Chrome, falling back automatically to gemini-3.5-flash-lite in the cloud.
  • Triage, Maintenance & Pattern Intelligence: Technicians and administrators manage repair workflows and trigger a 2nd-generation Cloud Function powered by Vertex AI (@google/genai with gemini-3.5-flash) to surface evidence-linked observations and unverified maintenance hypotheses across recurring incidents.

How we built it

  • Frontend: React, TypeScript, Vite, Tailwind CSS, Zustand, Zod, and IndexedDB (idb).
  • Backend & Security: Firebase Authentication, Cloud Firestore with strict organization-scoped security rules and status-transition guards, App Check (reCAPTCHA Enterprise), Firebase Hosting, and 2nd-generation Cloud Functions.
  • Hybrid & Server AI: Firebase AI Logic (PREFER_ON_DEVICE + cloud fallback) for individual incident extraction and @google/genai (Vertex AI) for server-side cross-incident pattern analysis.

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for MajiPulse

Share this project:

Updates

Submission history