Setting Up Voice Agents

This session explores how voice agents are configured, equipped with tools and tested within the Blue Dots approach. It covers prompt design, connecting agents to the tools they need, and the practical considerations involved in making conversations reliable across languages and dialects. The discussion draws on lessons from building voice agents for local job discovery. These include writing prompts that guide conversations without overwhelming the agent, using native-language scripts to improve pronunciation, avoiding invented information and ensuring agents never claim to have taken actions they have not performed. The session also demonstrates how an agent can be configured and tested, including a live test call in which the agent searched for jobs and applied on the user's behalf. It introduces automated test generation, scenario-based checks and the importance of manually reviewing calls, particularly during early iterations. Voice offers a way to uncover needs that people may not explicitly state through search or text. Building a reliable agent means continuously testing, learning from real conversations and improving how it responds. Explore the process of setting up a voice agent and the tools that help make it more reliable.