Dashboard Quick Reference¶

Accessing the Dashboard¶

# Start the dashboard
docker-compose up -d dashboard

# Open in browser
open http://localhost:8502

Dashboard Tabs¶

The dashboard is organized into three tabs:

Tab 1: 📊 Overview¶

Summary metrics and charts
System performance monitoring
Priority distribution

Tab 2: ⏳ Approval Queue ⭐ NEW¶

Human-in-the-Loop workflow management
Pending high-priority incidents (P1/P2)
Approve/Reject controls

Tab 3: 📋 All Incidents¶

Complete incident history
Searchable and filterable table
Detailed execution logs

Tab 1: Overview¶

1. Summary Metrics (Top Row)¶

Metric	Description
Total Incidents	All incidents processed in last 7 days
Critical (P1)	Priority 1 incidents requiring immediate response
High (P2)	Priority 2 incidents requiring response within 4 hours
Completed	Successfully processed incidents
Avg Time	Average processing time across all incidents

2. Charts¶

Incidents Over Time (24h) - Line chart showing incident volume by hour - Separate lines for Total, P1 Critical, P2 High - Hover for exact counts

Processing Time Trend - Bar chart of average processing time per hour - Color intensity shows performance (green = faster) - Identifies bottlenecks and slow periods

Priority Distribution - Donut chart showing P1/P2/P3/P4 breakdown - Color-coded: Red (P1), Orange (P2), Yellow (P3), Green (P4)

Status Overview - Bar chart: Completed, Failed, Processing - Shows system health and error rates

Tab 2: Approval Queue ⭐¶

Human-in-the-Loop Workflow Management

This tab demonstrates LangGraph's checkpoint-based HITL pattern. High-priority incidents (P1 Critical, P2 High) are automatically paused for human approval before spatial analysis.

What You'll See¶

Pending Approvals: Incidents awaiting human decision - Incident ID: Unique identifier - Severity Badge: 🔴 Critical, 🟠 High, 🟡 Medium, 🟢 Low - Time Waiting: Minutes since workflow paused - Incident Details: - Type, Location, Description - Reporter email, Urgency level - AI Classification (JSON)

Actions¶

✅ Approve & Process 1. Enter your name (e.g., "Emergency Manager Sarah Johnson") 2. Click "Approve & Process" 3. LangGraph workflow resumes from checkpoint 4. Spatial analysis and notifications proceed

❌ Reject 1. Enter your name 2. Provide rejection reason 3. Click "Reject" 4. Workflow terminates without further processing

How It Works¶

# When incident is submitted
if priority in ["P1", "P2"]:
    # Workflow pauses at interrupt point
    workflow.compile(interrupt_before=["spatial"])
    # State saved to MemorySaver checkpointer

# When human approves
workflow.update_state(config, {"human_approved": True})
workflow.invoke(None, config)  # Resume from checkpoint

Learn more: LangGraph HITL Pattern Guide

Tab 3: All Incidents¶

3. Recent Incidents Table¶

Columns: - Incident ID - Type (water_pollution, air_pollution, illegal_dumping, noise_pollution) - Location - Severity (critical, high, medium, low) - Priority (P1, P2, P3, P4) - Status (processing, completed, failed) - Processing Time - Steps Completed ✓ - Steps Failed ✗

Filters: - Priority: Select one or more (P1, P2, P3, P4) - Type: Filter by incident type - Status: Filter by processing status

4. Incident Details (Bottom Section)¶

Select any incident from the dropdown to view:

Execution Steps: Each step shows: - Status indicator (✅ Completed / ❌ Failed / 🔄 Processing) - Step name and order - Duration in milliseconds - Input data (JSON) - Output data (JSON) - Error message (if failed)

Workflow Steps: 1. Classify: Incident classification, severity, priority 2. Spatial: Protected sites, water bodies nearby 3. Guidance: Semantic search for relevant procedures 4. Notify: Email notifications sent

API Base URL: Change if API is on different host (default: http://localhost:8000)

Auto-refresh: Enable to refresh data automatically - Refresh interval: 5-60 seconds (slider) - Default: 10 seconds

Common Use Cases¶

Managing Human-in-the-Loop Approvals¶

Navigate to "Approval Queue" tab
Review pending P1/P2 incidents
Check classification and incident details
Approve to resume workflow or reject to terminate
View results in "All Incidents" tab

Monitoring Active Incidents¶

Enable auto-refresh (10s interval)
Watch the "Processing" count in Status Overview
Check Recent Incidents table for latest entries

Debugging Failed Incidents¶

Filter by Status: "failed"
Select failed incident from dropdown
Review Execution Steps for error messages
Check which step failed and why

Performance Analysis¶

Review "Processing Time Trend" chart
Identify hours with slow processing
Check "Avg Time" metric
Investigate specific slow incidents

Priority Monitoring¶

Watch P1/P2 metrics in top row
Review Priority Distribution chart
Filter table by P1 to see critical incidents
Ensure P1 incidents are processed quickly

API Integration¶

Dashboard calls these API endpoints:

# Summary stats
curl http://localhost:8000/api/v1/dashboard/summary

# Recent incidents (limit 50)
curl http://localhost:8000/api/v1/dashboard/incidents?limit=50

# Pending approvals (HITL)
curl http://localhost:8000/api/v1/incidents/pending-approval

# Approve incident
curl -X POST http://localhost:8000/api/v1/incidents/INC-XXX/approve \
  -H "Content-Type: application/json" \
  -d '{"approved_by": "Manager Name"}'

# Reject incident
curl -X POST http://localhost:8000/api/v1/incidents/INC-XXX/reject \
  -H "Content-Type: application/json" \
  -d '{"approved_by": "Manager Name", "reason": "Duplicate report"}'

# Specific incident logs
curl http://localhost:8000/api/v1/dashboard/incidents/INC-20231206-001/logs

# Hourly metrics (24 hours)
curl http://localhost:8000/api/v1/dashboard/metrics?hours=24

Tips & Tricks¶

Refresh Data: Click "Rerun" button in top-right or wait for auto-refresh

Full Screen: Click "Settings" → "Wide mode" for more screen space

Export Data: Hover over table → Click download icon → Export to CSV

Dark Mode: Settings → Theme → Dark (useful for monitoring displays)

Performance: Increase refresh interval if dashboard feels slow

Troubleshooting: Check API health at http://localhost:8000/health

Database Queries¶

Direct database queries for advanced analysis:

-- Recent execution logs
SELECT * FROM execution_logs 
ORDER BY created_at DESC 
LIMIT 100;

-- Failed steps today
SELECT incident_id, step_name, error_message 
FROM execution_logs 
WHERE status = 'failed' 
AND created_at > CURRENT_DATE;

-- Average processing time by step
SELECT 
    step_name,
    AVG(duration_ms) as avg_ms,
    COUNT(*) as count
FROM execution_logs
WHERE status = 'completed'
GROUP BY step_name;

-- Incidents by priority today
SELECT 
    classification->>'priority' as priority,
    COUNT(*) as count
FROM incidents
WHERE created_at > CURRENT_DATE
GROUP BY priority;

Keyboard Shortcuts¶

R - Rerun dashboard
C - Clear cache
S - Settings menu
/ - Search (in sidebar)

Support¶

For issues or questions: - Check API logs: docker logs defra-agent-api - Check dashboard logs: docker logs defra-dashboard - Review database: psql -h localhost -p 5433 -U postgres -d incident_reporting