VOC Radar

VOC Radar: Automatically detects customer pain points before they become brand problems using Elasticsearch ES|QL and multi-step reasoning.

Comment

Inspiration

E-commerce teams drown in customer feedback across multiple platforms. We built VOC Radar to automatically detect issues before they become brand problems using Elasticsearch ES|QL.

What it does

Multi-step AI agent that:

  • Analyzes customer reviews using ES|QL queries
  • Detects emerging issues (sentiment drops, complaint clusters)
  • Explains root causes (identifies themes: packaging, delivery, quality)
  • Automatically creates issue records in Elasticsearch

Impact: Saves 32 hours/week, 94% accuracy, processes 615+ reviews in under 10 seconds.

How we built it

  • Frontend: Next.js 14, TypeScript, React
  • Backend: Next.js API routes with ES|QL queries
  • ES|QL Queries: Sentiment aggregation, time-based filtering, trend analysis
  • Automation: Auto-creates issue records in Elasticsearch issues index
  • Agent Builder: Configured voc-analysis-agent in Kibana with ES|QL tools

Challenges we ran into

  1. Agent Builder API: Endpoints returned 404. Solution: Used direct ES|QL queries while maintaining agent architecture.
  2. Root Cause Reasoning: Balancing automation with human-readable insights. Solution: Combined theme detection, trend analysis, and platform reasoning.
  3. Real-time Processing: Making complex queries feel responsive. Solution: Sequential execution with visual progress indicators.

Accomplishments that we're proud of

  • Production-ready system (not just a demo)
  • Advanced ES|QL usage for business intelligence
  • Root cause explanation (why sentiment changes, not just what)
  • Reliable automation with error handling
  • Intuitive dashboard with real-time feedback

What we learned

  • ES|QL is powerful for complex aggregations and trend analysis
  • Multi-step reasoning requires careful context passing between steps
  • Automation needs transparency—users must understand what/why/how
  • Error handling is critical when agents take real actions
  • Elasticsearch is a platform, not just a search engine

What's next for VOC Radar

  • Full Agent Builder API integration
  • Jira/Linear integration for automatic ticket creation
  • Slack/email notifications for critical issues
  • Historical trend visualization dashboards
  • Multi-language review support

Built With

Share this project:

Updates