# Autonomous testing for a conversational voice agent

- Period: Jun 2026 – Jul 2026
- Domain: National telecom carrier
- Role: Architected the orchestrator-worker test framework; built the test-case design agent
- Key result: <10 min, to generate branch-covering test cases for a call flow
- Stack: Google ADK, Gemini, Claude API, LLM-as-judge, Python
- URL: https://farhanbaluch.com/work/voice-agent-testing

An orchestrator and parallel worker agents hold multi-turn conversations with a production voice assistant, while a design agent turns call-flow diagrams into test cases.

### Problem

Testing a conversational voice assistant by hand means calling it, over and over, for every branch of every call flow.

### What I built

- A test-case design agent that reads call-flow design documents and emits one test case per branch, reusing a library of sub-flows, checked against a 22-case gold standard.
- The orchestrator–worker framework (Google ADK and Gemini): an orchestrator dispatches parallel worker agents that run multi-turn conversations; responses are checked with semantic-equivalence scoring and reported through dashboards and alerts.