Refund_Datadog is a professional AI-driven application designed to streamline the process of requesting refunds for travel, hotels, and services. By leveraging a multi-agent architectural approach, it automates the journey from raw evidence extraction to generating professional legal-style refund letters.

• Multi-Agent AI Workflow:
  • Evidence Collector: Extracts receipts, transaction IDs, and merchant details from images, PDFs, and notes.
  • Policy Analyst: Cross-references user issues with airline/hotel policies (powered by Google Search integration).
  • Letter Generator: Produces professional, persuasive refund request letters tailored to the situation.
• Voice & Multimedia Support: Record voice notes or upload images/documents as evidence.
• Datadog Integration: Full Real User Monitoring (RUM), Session Replay, and error tracking for enterprise-grade stability.
• Privacy First: All case data is stored locally using IndexedDB. No personal documents are stored on a central server.
• Template System: Ready-to-use templates for common scenarios (Flight cancellation, Hotel hygiene, etc.).

• Frontend: React 18, Vite, TypeScript
• Styling: Tailwind CSS
• AI Core: Google Gemini SDK (gemini-3.0-flash)
• Monitoring: Datadog RUM, Datadog Logs, Session Replay
• Database: IndexedDB (Browser Local Storage)

This project is instrumented with Datadog for comprehensive monitoring:

• RUM (Real User Monitoring): Tracks user journeys and performance metrics.
• Session Replay: 20% sampled visual playback to debug UX issues.
• AI Performance: Custom metrics tracking latency and success rates of individual AI agents (refund.agent.extractEvidence, etc.).
• Error Tracking: Automated error reporting for both browser and AI service failures.

• Node.js (v18+)
• npm or yarn
• Google AI Studio API Key (Get it here)
• Datadog Account (Optional)

1. Clone the repository:

   git clone https://github.com/beiyuezou/Refund_Datadog.git
   cd Refund_Datadog

2. Install dependencies:

   npm install

3. Configure Environment Variables: Create a .env file in the root directory based on .env.example:

   cp .env.example .env

   Fill in your credentials:

   VITE_GEMINI_API_KEY=your_gemini_api_key
   VITE_DATADOG_APPLICATION_ID=your_id
   VITE_DATADOG_CLIENT_TOKEN=your_token
   VITE_DATADOG_SITE=us5.datadoghq.com

4. Launch Application:

   npm run dev

The system uses a sequential agent flow to ensure high accuracy:

1. Agent 1 (Extraction): Analyzes files/notes -> Returns structured JSON data.
2. Agent 2 (Reasoning): Uses the extracted data + Web Search -> Generates a policy-backed analysis.
3. Agent 3 (Synthesis): Uses the analysis -> Crafts the final professional letter.

This project is for educational and practical use. Please ensure you comply with merchant TOS when requesting refunds.