Genai Applied
Sign in Join the waitlist

Career track · Voice AI Developer

Become a voice AI developer: real-time speech in, speech out, for assistants and call agents.

Voice agents now hold natural conversations, handle interruptions and take actions mid-call. Learn to build them end to end, from the audio pipeline to the phone line, with latency users won't notice.

Or get your personal AI playbook now

What changes

Your work, re-mapped.

How AI is changing this work today, task by task. Your assessment maps it onto your own week.

Tasks AI can already do for you.

  • Rigid IVR phone menus
  • Manual call note-taking
  • Scripted first-line support calls
  • Hand-built speech pipelines

The curriculum

Ten modules.
One path through them.

Your assessment orders these for your level and goals and skips what you already know. Open any module to see what it covers.

01How voice AI works5 topics

The building blocks of a voice agent.

  1. Audio basics: sample rates and codecs
  2. Speech-to-text
  3. Text-to-speech
  4. Cascaded vs speech-to-speech
  5. The latency budget
02Speech recognition in practice5 topics

Get accurate transcripts in real conditions.

  1. Streaming vs batch STT
  2. Accents, noise and domain terms
  3. Custom vocabulary
  4. Measuring word error rate
  5. Multilingual speech
03Voices and speech synthesis5 topics

Make agents sound natural and on-brand.

  1. TTS engines compared
  2. Streaming synthesis
  3. Voice selection and cloning ethics
  4. Prosody and pronunciation
  5. Cost per minute
04Speech-to-speech models5 topics

Native audio models such as Gemini Live and OpenAI Realtime.

  1. How speech-to-speech works
  2. Sessions and context limits
  3. Tool calling mid-conversation
  4. Ephemeral tokens and security
  5. Trade-offs vs cascaded pipelines
05Turn-taking and interruptions5 topics

The difference between a demo and a natural conversation.

  1. Voice activity detection
  2. Semantic turn detection
  3. Barge-in handling
  4. Echo cancellation
  5. Backchannels and silence
06Voice agent frameworks5 topics

Build pipelines with Pipecat and LiveKit.

  1. Pipeline architecture
  2. WebRTC transport
  3. Pipecat in practice
  4. LiveKit Agents in practice
  5. Choosing a framework
07Telephony5 topics

Put your agent on a phone number.

  1. SIP and PSTN basics
  2. Twilio and telephony providers
  3. Call transfer to humans
  4. Recording and consent
  5. Outbound calling rules
08Conversation design and tools5 topics

Agents that complete tasks, not just chat.

  1. Designing voice prompts
  2. Tools: bookings, lookups, payments
  3. Confirmations and error recovery
  4. Persona and tone
  5. Multilingual conversations
09Production voice systems5 topics

Scale, monitor and keep quality high.

  1. Concurrency and scaling
  2. Latency monitoring
  3. Call analytics and QA
  4. Testing voice agents
  5. Cost control
10Capstone: a production voice agent5 topics

Build a voice agent that handles a real use case end to end.

  1. Use case and success metrics
  2. Build and integrate tools
  3. Phone or web deployment
  4. Evaluation with real calls
  5. Demo and portfolio

Outcomes

By the end, you can.

  • 01Build a real-time voice agent with natural turn-taking
  • 02Choose between speech-to-speech and cascaded pipelines
  • 03Connect a voice agent to a phone line with human handoff
  • 04Monitor latency, quality and cost in production

Tools and techniques you'll use

  • Gemini Live API
  • OpenAI Realtime API
  • Pipecat
  • LiveKit Agents
  • Deepgram
  • ElevenLabs
  • Silero VAD
  • Twilio and SIP

Who it's forDevelopers and AI engineers building voice assistants, call-centre agents and voice features.

The course is opening soon

Be first in.

Join the waitlist for the Voice AI Developer course. One email when it opens, nothing else.

Questions

Good to know.

Speech-to-speech or STT-LLM-TTS: which is better?

Speech-to-speech models such as Gemini Live and OpenAI Realtime keep tone and timing and are simpler to wire up, while cascaded pipelines give more control over each step. The course builds both so you can choose per use case.

What latency should a voice agent have?

Responses that start within about half a second to a second feel natural. Handling interruptions well matters as much as raw speed, and the course covers both.

When does the course open?

The full course is being built now and opens to the waitlist first. Join with your email and we'll tell you as soon as it's ready.

How is the course personalised?

Before the course, a five-minute assessment, by text or by voice with Iris, our AI guide, learns your role, daily work, skills and goals. Your personal playbook then orders these modules for your level and goals and skips what you already know.

What can I do while I wait?

Take the assessment now. You get your personal AI playbook straight away, with the tasks in your work AI can take over, the ones it speeds up, and your learning path.