← All case studies

Multilingual testing for insurance voice agents

Strala connects its LiveKit agents to Roark and runs simulated callers in five languages, with 54 personas and more than 870 test flows.

54
personas
870+
test flows
Over 80%
of runs triggered through the Roark API
We've been using Roark extensively. It's been really helpful for testing different scenarios, foreign languages, and self-improve our agents.

Armando Schmid

Co-Founder and CTO, Strala

Challenge

Strala's inbound agents serve insurance customers, so what they can and cannot say is tightly controlled, and every branch of a conversation needs to stay grounded in the agent's instructions. Before Roark, there was no scalable way to test call scenarios end to end, so every prompt or backend change had to be re-checked across many flows and languages. Tool calls also have to be filled in accurately on every path.

Approach

Strala connects its LiveKit agents to Roark and runs simulated callers in five languages, with 54 personas and more than 870 test flows. Most runs (over 80%) are triggered automatically through the Roark API. Each simulated call is checked against Strala's own expectations for how the agent should behave, and production calls are monitored in the same platform and proactively flagged before there's any potential model drift.

Results

  • Automated multilingual simulation testing across five languages, replacing manual call QA
  • Clear improvement in how often agents meet Strala's own expectations on the same test suites
  • Most simulation runs are triggered automatically through the Roark API, so prompt and backend changes can be re-checked across many flows and languages
  • Production call monitoring in the same platform, with proactive drift flagging

Test your agents the way Strala does.

Roark is a voice and chat agent observability and simulation platform. Book a demo and we will run it on your own agent.