Building advanced conversational AI now includes real-time, voice-driven agents. These require low-latency, two-way communication and real-time information retrieval. This guide demonstrates building such an agent using Gemini and the Google Agent Development Kit (ADK). The process starts with a foundational agent possessing a persona but no external tool access. This basic agent can chat using its pre-trained knowledge. To enhance functionality, advanced agents are equipped with tools for accessing live data and services. The ADK provides tools like Google Search for real-time information and MCP Toolset for Google Maps integration. The agent acts as an orchestrator, delegating tasks to specialized tools. The RunConfig object manages communication, including voice selection and bidirectional streaming. Bidirectional streaming allows users to interrupt the agent, fostering more natural conversations. Real-time voice chats are managed asynchronously using Python's asyncio and TaskGroup to handle listening, thinking, and speaking concurrently. Audio data is transmitted using Base64 encoding, while text transcriptions are streamed for immediate user feedback. Resources include the Google ADK, Gemini's Native Audio, and project source code on GitHub.