Logo
AgentDune Voice

AgentDune Voice

A production-ready voice AI demonstration platform built for ARASCO, one of the Gulf region largest agribusiness companies. Leverages OpenAI Realtime API with WebRTC for sub-second voice interactions. Features 6 specialized agent personas with 30+ callable tools for order tracking, inventory checks, feed calculations, and bilingual Arabic/English support.

Visit Project
Next.js 15 OpenAI Realtime API WebRTC Whisper Framer Motion

Features

  • Sub-second latency voice interactions using WebRTC peer connections
  • 6 specialized agent personas: customer service, order tracking, agricultural support, food safety lab, veterinary health, retail assistance
  • 30+ callable enterprise tools for order lookup, inventory, dosage calculations, recipes
  • Bilingual support with native Arabic and English responses including Saudi dialect
  • Live transcription with Whisper and interim/final transcription states
  • Tool result visualization alongside agent controls
  • Multiple TTS voice options (ballad, ash, sage, nova)
  • Real-time token usage monitoring

Challenge

Building a voice AI system capable of handling complex multi-domain customer interactions in real-time for a large agribusiness, with native Arabic support and integration with enterprise systems for order tracking, inventory, and specialized calculations.

Solution

Implemented OpenAI’s Realtime API with WebRTC for sub-second latency. Created 6 specialized agent personas each with domain-specific tool sets. Built a sophisticated tool-calling architecture allowing real-time function execution during conversations while maintaining conversation state.

Highlights

  • Sub-second voice latency with WebRTC
  • 30+ enterprise tools callable during live conversations
  • Native Arabic support with Saudi dialect