Refund_Datadog

Refund_Datadog, now monitored via Datadog, streamlines travel refunds for flights and hotels. We focus on securing the refunds consumers deserve, making travel-related disputes hassle-free.

Comment

Inspiration

As our multi-agent application, Refund, prepares for high-traffic production environments, robust monitoring became a critical necessity. We needed a way to gain real-time visibility into agent behaviors and system performance to proactively identify and resolve bottlenecks before they impact our users.

What it does

Refund_Datadog serves as the vital bridge between the Refund application and the Datadog observability platform. It ingests, aggregates, and analyzes application data, system logs, and user interaction patterns, providing a centralized dashboard for the entire multi-agent ecosystem.

How we built it

Core Logic: Prototyped in Google AI Studio for rapid iteration. Development Framework: Deeply integrated using Google Antigravity for complex agent orchestration. Observability: Seamlessly integrated Datadog RUM (Real User Monitoring) and Log Management. Models: Powered by a hybrid architecture of Gemini 3.0 Flash (for speed) and Claude 4.5 (for complex reasoning).

Challenges we ran into

The primary hurdle was the steep learning curve of the Datadog ecosystem. Navigating the vast UI to locate specific application entry points and configuring custom dashboards for multi-agent workflows proved challenging during the initial setup.

that we're proud of

We mastered Datadog’s powerful Global Search and query syntax, which acted as a "skeleton key" to unlocking the full potential of the platform. This allowed us to efficiently locate hidden components and visualize data flows that were previously obscured.

What we learned

We gained deep expertise in implementing End-to-End Observability. We now understand how to leverage telemetry to ensure the long-term health, reliability, and sustainable growth of an AI-driven application.

What's next for Refund_Datadog

Logic Refinement: Adapting monitoring schemas as business logic evolves. Deep-Dive Scenarios: Stress-testing specific edge cases in refund disputes. User Testing: Conducting live user trials to correlate technical metrics with actual user satisfaction.

Built With

  • antigravity
  • claude4.5
  • gemini3flash
  • googleaistudio
  • react
Share this project:

Updates