Module 1 · Voice Agent Foundations

OBJECTIVE: understand how an AI phone/voice agent works from the moment someone speaks to the moment it replies — and every building block you will configure on Vapi or code with LiveKit. WHAT WE COVER: • What voice agents are; inbound vs outbound calling; where businesses use them • The voice pipeline: speech-to-text (STT) → LLM → text-to-speech (TTS), and speech-to-speech models • Audio basics you need: sample rate, codecs, phone audio vs web audio • Real-time conversation: streaming, voice activity detection (VAD), turn detection, barge-in/interruptions, latency budget • Phone basics: phone numbers, PSTN, SIP and SIP trunks, call transfer, DTMF keypad input, voicemail • The LLM part for voice: writing prompts for speech, tool calling, knowledge (RAG), memory and state • Choosing providers: STT, LLM, TTS, telephony — including Indian languages • Rules: consent, recording disclosure, do-not-disturb, personal data • Two ways to build: plug-and-play platforms (Vapi, Retell) vs your own (LiveKit, Pipecat) AFTER THIS MODULE YOU CAN: • Draw the full voice-agent pipeline and explain each block • Explain why latency and turn-taking decide whether callers like the agent • Choose STT, LLM, TTS and telephony providers for a use case and budget PREREQUISITES: basic Python and calling an API. If you followed the full path, the 'Already covered' column tells you what to skip.