Advanced

Capstone: Production Voice Agent E2E — 6-8 hours

Build a complete production voice assistant from scratch: real-time Whisper transcription, Claude reasoning with conversation memory, ElevenLabs streaming TTS, and latency optimization. Three working examples you can deploy the same day.

Free — Course included in Voice Agents formation

S'abonner — 9,99 €/mois

6-8 hours

7h of training

6 modules

Max 12 participants

VoiceWhisperClaudeElevenLabsWebSocketStreamingPython

Free

Included with Voice Agents formation

Next session: On request

S'abonner — 9,99 €/mois Request group pricing

✓Official completion certificate

✓Lifetime access to resources

✓Post-training instructor support

AI Act : la formation IA devient obligatoire avant le 2 aout 2026

Anticipez la mise en conformite de votre entreprise. Voir nos formations

What you will build and learn

Skills you can apply in production the same day

✓Transcribe speech in real-time with Whisper using VAD and vocabulary hints
✓Stream Claude responses to TTS in under 900ms time-to-first-token
✓Detect sentence boundaries for gapless audio playback
✓Budget and track voice pipeline costs (target: <$0.006/turn)
✓Implement graceful error handling and human escalation
✓Deploy a complete voice agent with FastAPI WebSocket backend

Detailed curriculum

6 modules · 7h of intensive hands-on training

01Architecture & Latency Budget

1h00

●Latency budget calculator
●STT/LLM/TTS trade-offs
●Cost modeling ($0.006/turn)

02Whisper STT Streaming

1h30

●Browser WebSocket audio capture
●FastAPI transcription server
●VAD and vocabulary injection

03Claude Context-Aware Reasoning

1h30

●Streaming with TTFT measurement
●Sentence boundary detection
●Context injection patterns

04ElevenLabs TTS Streaming

1h00

●Streaming to speakers (<350ms)
●Browser MediaSource API
●Response caching strategy

05E2E Customer Service Chatbot

2h00

●Full FastAPI WebSocket server
●3-service integration
●Escalation detection + grading rubric

06Advanced Examples

2h00

●Technical support agent with tool calling
●Voice note processor
●Production deployment checklist

Who is this course for?

Target audience

✓Intermediate Developers

✓Full-Stack Engineers

✓AI/ML Engineers

Advanced7h · 12 participants max

Prerequisites

●Python 3.11+ and asyncio fundamentals
●REST API consumption (requests, httpx, or fetch)
●Basic understanding of WebSockets
●Completed 'Voice Agents in Production' module or equivalent experience

Format

Online

Duration

6-8 hours (7h)

Next session

On request

Certification

Certificate of Completion

Frequently asked questions

Everything you need to know before enrolling

Do I need to have taken the Voice Agents formation first?

Recommended but not required. You need Python async fundamentals and basic WebSocket knowledge. If you can build a FastAPI endpoint, you're ready.

What API costs will I incur during the exercises?

Approximately $0.50–2 for the full capstone with all examples, using real API calls. Whisper, Claude Haiku, and ElevenLabs each have pay-as-you-go pricing with no minimum.

Can I use this code in production?

Yes. The code is MIT-licensed and production-ready. The E2E example has been deployed to AWS Lambda + API Gateway WebSocket in real projects.

Is there a French version?

Yes — the French version is available at /formations/agent-vocal-capstone.