How I used AI voice to help me play Magic: The Gathering
summary
Built an AI voice assistant that answers Magic: The Gathering rules questions by piping Twilio voice through a websocket to a Python app, calling Twilio Conversation Relay, an LLM (configured here as OpenAI gpt-4o-mini), and 11Labs TTS; session state preserves context so follow-ups work. A UK number is provisioned, Enro exposes localhost, and the system prints transcripts to console. Live demo struggles with conference mic but a prerecorded call shows Jace resolving mechanics, handling interruptions, and returning spoken clarifications reliably.
AI-written summary · about
chapters (10)
- 0:00Welcome
- 1:33Human vs computer dev
- 10:29Voice AI frontier
- 18:19Voice AI challenges
- 25:22Voice AI real-world use
- 29:55MTG AI assistant
- 48:13Live demo troubleshooting
- 55:44System architecture
- 1:00:01Future of AI voice
- 1:03:09Key takeaways
related talks
51mBuilding an AI assistant that makes phone calls
01h 02m[Full Workshop] Building Conversational AI Agents
01h 58mWhat If Your Backend Could Talk? Building Real-Time AI Voice Agents
01h 04m“Hello, AI?!” – Real-Time Voice Interactions with LLMs
01h 26mBuilding voice agents with OpenAI
23mBuilding with OpenAI: What's Ahead
17mBuilding Effective Voice Agents
01h 00mCalling AI Tools from Java