Voice AI agent course Build a real-time voice agent
Build an AI receptionist that answers calls, lets callers interrupt, checks a calendar mid-conversation and books appointments only after a clear yes, on free tiers and free local tools.
One-time payment · about US$17 · international cards accepted
7 video lessons (3 h 22 min) · 7 coding rounds · capstone · certificate
Enrol nowLevelIntermediate
You needPython, async basics
Modules7, in order
Ends withCertificate
Taught by Team Homingo
What you’ll build: a voice agent callers can interrupt
01A real-time pipeline: 20 ms audio frames, endpointing, speech to text, a streaming model and text to speech, measured per stage
02Voice activity detection with hysteresis, pre-roll and a tuned end-of-turn timeout
03Streaming replies: the first clause is spoken while the model is still writing
04Barge-in: the agent stops about half a second after the caller starts talking, and the history keeps only what the caller heard
05Calendar tools mid-call: free slots, holds, idempotent bookings and a confirmation guard in code
06Phone audio (G.711 mu-law), call flow, safety guards, an eval gate, and a live demo on a free tier
What you get
- 7 modules of lessons
- From the real-time voice loop and its latency budget to safety, evaluation and shipping the ai receptionist, in order.
- A project in every module
- Each module ends with something that runs.
- Capstone project
- Your own project, in a public GitHub repo with a live demo.
- Certificate of completion
- From Homingo, with your name, when you finish.
- Help
- By email at official@homingo.co.in
Syllabus
7 modules, each ending in code that runs.
Every module ends with a coding round: one more working part of the project, written by you and checked before you move on.
#ModuleYou build
0The real-time voice loop and its latency budgetThe voicedesk repo running end to end in offline mode, and a turn-metrics function that splits every turn into endpointing, STT, model, TTS and playout time against a voice-to-voice budget
- 0.1The real-time voice loop: where every millisecond goes22:00
- 0.2Check: the voice loop and its budget05:00
- 0.3Coding round: per-stage turn metrics15:00
1Audio in: frames, voice activity and endpointingAn endpointer that hears the caller: frame loudness in dBFS, a start threshold and a lower end threshold (hysteresis), a minimum speech length, and a silence timeout that ends the turn (the repo adds 200 ms of pre-roll around it, so the first syllable survives)
2Streaming speech: transcripts in, speakable chunks outA speech chunker that turns the model's token stream into speakable chunks (an early first clause, whole sentences after, no breaks after Dr. or inside 10.30), spoken-form times, and the free STT and TTS adapters wired in
3Turn-taking and interruptionsBarge-in that works: three conversation states, an interruption rule that ignores clicks and brief noises, agent audio that stops about half a second after the caller starts talking, cancelled model and TTS work, and a history cut to exactly what the caller heard
4Tool calls mid-conversation: calendar and bookingCalendar tools the agent calls while it talks: free-slot search over opening hours and busy intervals, a hold-then-book flow that cannot double-book, idempotent retries, a booking guard that needs the caller's own yes, and a filler line when a tool is slow
5Call handling: phone lines, call flow and hand-offsThe call around the conversation: you write a G.711 mu-law codec for phone audio and a silence ladder that re-prompts, then hangs up. The repo provides the rest around them: 8 kHz resampling, an optional Twilio Media Streams line, the AI and recording notice, opening hours, a call-length ceiling and the call summary
6Safety, evaluation and shipping the AI receptionistThe finished AI receptionist: masked phone numbers and emails in every stored transcript, bookings only for the caller's own number (or, without caller id, only those made on the same call), emergencies answered without the model, prompt injections that change nothing, an eval gate over ten scripted calls, and a public repo with a live demo
Certificate
Finish the work, get the certificate.
Complete every module and the capstone project, and you receive a certificate of completion from Homingo with your name and the course title.
Your instructor
Team Homingo
Wrote every module and coding round of this course, and answers your questions by email.
FAQ
Questions before you enrol
- A capstone project in your own public GitHub repo with a live demo, and a certificate of completion from Homingo: work you can show and run, not just notes.