Roark

Roark assists teams in testing, monitoring, and enhancing their voice agents. As the demand for Voice AI grows, ensuring reliability remains a significant challenge. Teams often spend extensive hours manually testing agents, yet failures can still occur. In the past six months, Roark has processed over 10 million minutes of calls, supporting monitoring and simulation for teams within the Voice AI ecosystem, including those from YC companies. Roark addresses these challenges by offering: 📊 Monitoring & Evaluation: Features over 40 built-in metrics such as latency, instruction-follow, and sentiment analysis, along with custom dashboards, alerts, and the capability to define personalized metrics. It supports up to 15 speakers with automatic speaker identification. 🎭 Simulations & Personas: Provides end-to-end phone and WebSocket simulations for both inbound and outbound agents, with configurable personas that include accents, languages, and speech and behavior profiles. Tests can be defined as conversations using a graph-based approach, allowing easy branching into edge cases and variants. 🔁 Full Lifecycle Loop: Automatically converts failed calls into repeatable tests, ensuring that each failure contributes to strengthening the agent. Roark serves as the essential QA layer for Voice AI, enabling teams to deploy agents with confidence.