Tiger AI Voice Onboarding Prototype

This repository contains two deliverables:

1. A working Next.js prototype for the Tiger Credit Card AI voice onboarding operations system.
2. A production-ready architecture scaffold for a real voice agent.

Prototype features

• Overview dashboard
• Customer queue
• Customer detail / journey view
• Agent playground with both Offline (Rule-based Regex) and Online (Google Gemini 1.5 Flash LLM) voice modes
• Golden test evaluation page
• Interactive architecture diagram and presentation views
• Embedded reference documents (Codex Spec, Prompt Blueprint, Agent Mega-Prompt)
• Deterministic stage and next-best-action logic in pure functions
• Seeded data with browser persistence

Run locally

npm install
npm run dev

Then open http://localhost:3000.

Build

npm run build

Demo storyline

1. Open Overview and explain funnel drop-off.
2. Open Priya in Customer Queue.
3. Show EKYC_PENDING -> GUIDE_EKYC.
4. Open Agent Playground and click Ye KYC bahut complicated hai.
5. Show objection detection and Hinglish response.
6. Ask for WhatsApp and review the tool call.
7. Change Priya eKYC to COMPLETED in Customer Detail.
8. Refresh and show VKYC_PENDING.
9. Open Evaluations and click Run All Tests.
10. Review guardrails such as OTP handling and failed-tool behavior.

Production-ready voice agent scaffold

See ../voice-agent-production/ for a production architecture package with:

• domain orchestration layer
• provider adapter interfaces
• policy and guardrail services
• telephony, ASR, TTS, and WhatsApp boundaries
• observability and deployment configuration examples