Voice-Driven B2B Lead Generation Agent
A voice-first LiveKit agent that collects, confirms, and delivers B2B leads through natural conversation.
SWE (AI/MLOps)Voice AILiveKitLLM AgentsSpeech-to-TextText-to-Speech
A real-time voice agent built on LiveKit's agent SDK, using Google for speech-to-text and text-to-speech and a Tavus video avatar for the conversational interface. The agent walks a caller through a four-step flow, collecting target industries or business names, requesting an email address, reading it back for spoken confirmation, and forwarding the structured lead (industry, business, email, timestamp) to a webhook for downstream CRM processing, all within a single real-time voice session with noise cancellation.
What I built
- Built a real-time voice agent on LiveKit's agent SDK, wiring together Google speech-to-text and text-to-speech and a Tavus video avatar into a single conversational session.
- Implemented a four-step lead-capture flow: collect industry and business names, request an email address, read it back for spoken confirmation, then submit the structured lead to a webhook.
- Built an industry-extraction step to normalize free-form spoken business and industry mentions into clean structured fields before submission.
- Used aiohttp for asynchronous webhook delivery, disconnecting the session only after the lead payload (industry, business, email, timestamp) is confirmed sent.
- Added noise cancellation via LiveKit plugins to keep transcription reliable under real conversational audio conditions.
Tech stack
SWE (AI/MLOps)Voice AILiveKitLLM AgentsSpeech-to-TextText-to-Speech