← Back to Our Work

AI Dungeons & Dragons

Voice-driven AI tabletop: an AI Dungeon Master and autonomous AI companions with real-time talking-head avatars

Real-Time Voice / Multi-Agent AI
View Live Site

Project Summary

A production AI D&D experience at gurucloudai.com/dnd: a human player speaks their actions aloud, an AI Dungeon Master narrates the world, and a party of autonomous AI companions converse with the player and each other in real time. Characters speak with emotion-directed synthesized voices over animated portraits, or as live WebRTC talking-head avatars. Under the hood it is a latency-engineered multi-agent pipeline: a fast narration pass returns instantly while an authoritative stronger-model adjudication of game state runs asynchronously off the critical path, and a turn-taking broker arbitrates which companion speaks each beat.

Our role: Full product: architecture, engineering, voice pipeline

What We Did

  • Voice-in / voice-out play loop in production: spoken player actions transcribed inline (Whisper), character lines rendered as emotion-directed speech (12 delivery cues) or live WebRTC talking heads
  • Fast-turn architecture: player-facing narration returns inline while authoritative game-state adjudication runs on a stronger model asynchronously — per-call latency_ms instrumented and recorded
  • Round-broker turn arbitration: companions propose lines in parallel with self-reported urgency, the winner's line feeds back so remaining companions react to it — bounded cost per beat, no crosstalk
  • Live avatars are strictly script-driven: the vendor's own LLM is locked out and input audio disabled so nothing but our DM can make a character speak

Want something like this?

Describe what you'd build and get a scoped plan with a budget — drafted in minutes, confirmed by a human.

Scope Your Project — Free