Building Conversational Agents
summary
Shows how to build tool-using coding agents with Google DeepMind Gemini APIs: a coding agent that calls JSON-schema read/write/run tools, writes an SVG/CSV, and executes bash; Interactions API provides server-side state with previous-interaction IDs, improving caching (seen 2–3×) and offering async deep-research agents. Live demos use Gemini 3.1 Flash Live + Lyra 3 for realtime voice and music generation, explain WebSocket/ephemeral-token setups, session limits (audio 15 min, A/V 2 min) and context compression — but Google Search grounding in the Live demo fails intermittently.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:14Introduction and speaker introductions
- 6:15Audience interaction and project discussions
- 8:38Introduction to building conversational agents
- 28:17Discussion on Gemini Flash for coding and agentic use
- 36:28Coding agent implementation and tool calling demonstration
- 42:55Overview of the Interactions API and state management
- 49:05Introduction to the Gemini Live API
- 50:02Live Jukebox demo with music generation
- 54:49Deep dive into Gemini Flash Live features (multimodality, latency, tools)
- 1:06:54Technical setup and implementation of the Live API using WebSockets
- 1:25:14Session management and context window compression
- 1:26:57Real-world business use cases for conversational agents
- 1:35:02Multimodal grounding and handling audio inputs
- 1:40:00Discussion on personalization and speaker identification
related talks
39mBuild the new generation of AI Agents
01h 45mAI Engineering with the Google Gemini 2.5 Model Family
37mBuild agents with Gemini API
49mLet's build an AI agent
01h 00mHow to work with generative AI in JavaScript
50mLet's build an AI agent
15mBuilding with the Gemini API and AI Studio
45mAgentic Dev: Stitch and Jules in the Antigravity of the Gemini Constellation