Engineering PerciBot: DDD + TDD for Agentic Voice AI

This session walks through the implementation of PerciBot, a platform that uses LangGraph agents and ElevenLabs realtime voice to handle payment negotiations. The architecture follows Domain-Driven Design (DDD), keeping rules and entities pure, while agents act as clients of the same use cases.

With ElevenLabs, the agent can hold real-time conversations, capture transcripts, and route them through a tool-calling workflow: classify intent, record outcomes, or trigger payments. Every side effect goes through the application layer, ensuring auditability. We developed everything with Test-Driven Development (TDD), unit tests for policies, contract tests for repos, and integration tests for agent flows, to keep the system reliable as it evolved.