Hamza .

01

AI project

Customer-service AI agent for a 500k-customer highway operator

A production multilingual (English + Malay) LangGraph multi-agent assistant serving 500k+ customers of a Malaysian highway operator across chat and voice. It delegates between specialist agents, validates every tool call with Pydantic, and hands off to human live-chat agents when it should. I added a Gemini Live voice agent over WebRTC and SIP telephony (a Go SIP/RTP worker load-tested at 45 concurrent calls), real-time call transcription with auto-reconnect, response caching that cut latency, and a multi-tenant AI gateway with per-organisation RBAC, guardrails and RAG.

  • LangGraph,
  • Python,
  • Gemini Live,
  • Go,
  • WebRTC / SIP,
  • Rails,
  • Redis,
  • GCP / Azure