AI Decision Copilot for Faster Community Operations

Turn scattered live community signals into prioritized incidents, recommended actions, and faster response decisions.

Gen AI Academy APAC Edition Cohort 2 — AI for Better Living and Smarter Communities

CivicPulse is a data intelligence tool that helps community operators make faster and better decisions by unifying live signals, prioritizing incidents with AI-powered scoring, and recommending the next best action.

Community operations teams — housing societies, NGOs, campus facilities, local administrators — face fragmented signals across multiple sources (manual reports, weather feeds, facility status, citizen complaints). Prioritization under pressure is hard. An operator receiving 15+ simultaneous reports during a heatwave must decide: which issue threatens the most people? Which one is compounding? Which team should respond first?

CivicPulse ingests all signals, normalizes them into a common model, enriches them with AI context, ranks them with a transparent 10-factor scoring engine, and generates actionable recommendations and handoff summaries — all in seconds.

Core workflow: ingest → enrich → prioritize → recommend → explain → hand off

CivicPulse uses a 5-agent pipeline:

1. Intake & Normalization Agent — Receives raw signals from 7+ sources, normalizes into a common incident model, detects duplicate clusters via spatial + temporal proximity
2. Context Enrichment Agent — Uses NVIDIA NIM to analyze weather context, proximity to other incidents, compounding risks, and assign recommended response teams
3. Priority Scoring Agent — Scores incidents on 10 factors (urgency, severity, population impact, compounding risk, time sensitivity, resource constraints, location context, signal confidence, duplicate clustering, service criticality) and produces explainable rankings
4. Action Recommendation Agent — Generates immediate next steps, 30-minute action plans, required resources, safety notes, and 24-hour risk assessments
5. Communications / Handoff Agent — Produces operator handoff summaries, field messages (WhatsApp/SMS style), supervisor escalation notes, and public update drafts

The Decision Acceleration Panel compares manual vs AI-assisted metrics:

• Google Cloud Run — Primary deployment target (free tier: 2M requests/mo)
• Google Artifact Registry — Container image storage
• NVIDIA NIM API — All AI inference (meta/llama-3.1-8b-instruct)

Cost commitment: Runs entirely within Google Cloud free tier. No billing required.

• All AI inference through NVIDIA NIM API (integrate.api.nvidia.com)
• Model: meta/llama-3.1-8b-instruct
• Structured JSON outputs with schema validation
• Retry logic with fallback to deterministic scoring
• Used for: context enrichment, recommendation generation, explanation generation, communications drafting

• Node.js 20+
• npm
• NVIDIA NIM API key (get from build.nvidia.com)

# Clone the repository
git clone https://github.com/your-username/civicpulse.git
cd civicpulse

# Install dependencies
npm install

# Set up environment
cp .env.example .env
# Edit .env and add your NVIDIA_API_KEY

# Run development server
npm run dev

# Open http://localhost:3000

# Build the Docker image
docker build -t gcr.io/YOUR_PROJECT_ID/civicpulse .

# Push to Artifact Registry
docker push gcr.io/YOUR_PROJECT_ID/civicpulse

# Deploy to Cloud Run
gcloud run deploy civicpulse \
  --image gcr.io/YOUR_PROJECT_ID/civicpulse \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --set-env-vars NVIDIA_API_KEY=your-key-here

Or using source-based deployment:

gcloud run deploy civicpulse --source . --platform managed --region us-central1

1. Open the app → Landing page with value proposition
2. Click "Try Demo" → Select a scenario (Heatwave + Water Shortage, Flooding + Traffic, or Clinic Supply Shortage)
3. Watch the AI pipeline process: normalize → enrich → score → recommend
4. Browse the ranked priority queue with severity/urgency badges
5. Click the top incident → See why it was prioritized, score breakdown, and AI enrichment
6. Switch to "Actions & Plan" tab → See immediate next step, resources, safety notes, 30-min plan
7. Click "Generate Handoff" → Get operator handoff, field message, escalation note, public update
8. Go to "Ask CivicPulse" → Ask questions grounded in system state
9. View Architecture page → See full agent pipeline and Google Cloud usage

• Demo mode uses seeded synthetic data (not live feeds)
• AI enrichment falls back to deterministic scoring when NIM API is unavailable
• No persistent database — state is session-based
• Single-region deployment (no multi-region failover)

• Real-time data feeds via Pub/Sub and Cloud Scheduler
• BigQuery integration for historical analytics
• Multi-tenant support for different organizations
• Mobile-responsive field interface
• Integration with real weather and traffic APIs
• Google ADK integration for production agent orchestration
• Push notifications for critical incidents

• Frontend: Next.js 15, React 19, TypeScript, Tailwind CSS v4, shadcn/ui, lucide-react
• Validation: Zod
• AI: NVIDIA NIM (meta/llama-3.1-8b-instruct)
• Deployment: Google Cloud Run, Docker
• CI/CD: GitHub Actions

Built for Gen AI Academy APAC Edition Cohort 2 Hackathon