Loading video player...
Learn how to build a sophisticated real-time voice sales agent that can have natural conversations with potential customers. You'll create both single-agent and multi-agent systems where specialized AI assistants handle sales, technical support, and pricing inquiries. https://twitter.com/SarahChieng Timestamps 0:21 Introduction to the Workshop & Sales Agent Concept 1:18 Workshop Benefits & Starter Code 2:01 Introduction to Cerebras and its Hardware 3:34 Cerebras Hardware Innovations (Memory Bandwidth) 6:56 Software Innovations (Speculative Decoding) 8:10 Recap of Hardware and Software Optimizations 8:42 The Power of Voice Agents in Customer Interaction 9:57 Key Capabilities of Voice Agents 10:24 Applications of Voice Agents (Customer Service, Sales, Tech Support) 10:39 Inside a Voice Agent: Breaking Down the Conversation 10:46 Phase 1: Listening (Speech to Text & End of Utterance Detection) 11:38 Phase 2: Thinking (Large Language Model & Context Pulling) 12:04 Phase 3: Speaking (Text to Speech & Real-time Responses) 12:26 LiveKit Agent SDK for Orchestration 12:56 Installing Necessary Packages (LiveKit, Cartesia, OpenAI) 13:21 Cerebras Speed Comparison 14:07 Installing LiveKit CLI (Optional) 14:50 Understanding LiveKit and WebRTC 15:44 LiveKit Applications (Meetings, Call Centers, Websites) 16:17 Integrating Cerebras, LiveKit, and Cartesia 16:32 Step 2: Teaching the AI Sales Agent about the Business 16:45 Minimizing Hallucinations with Custom Context 17:10 Loading Business Information (Pricing, Product Descriptions, Objections) 17:58 Structured Format for AI Context 18:13 Step 3: Creating the Sales Agent (Code Walkthrough) 18:54 Implementing the Sales Agent Class (Prompting, Context, Initialization) 19:48 Defining the on_enter Method for Conversation Start 20:11 Step 4: Launching and Running the Agent 20:55 Expanding the Sales Agent: Multi-Agent Systems 21:16 Types of Specialized Agents (Greeting, Sales, Technical, Pricing) 22:10 Agent Handoff in Multi-Agent Systems 22:37 Adding Tool Calling 22:50 Conclusion & Q&A