Building the next interface - AI personas who feel as natural as interacting with a human
summary
Company launches beta with over 300 customers using personas built from WebRTC → VAD → transcription → LLM → TTS → animation → renderer, claiming full‑pixel generation rather than lip‑dub. Engineering reports 744 ms average latency versus an 800 millisecond northstar and over 1,000 failed experiments tracked while targeting subsecond responsiveness. Demo attempts include a live browser embed that stumbles on audio and HTML errors, and a oneshot persona feature that hallucinates missing facial detail to produce smooth real‑time video.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:25Live demo: AI persona interaction
- 3:53Why AI personas matter
- 8:57Science of natural AI faces
- 9:18Technical challenges in personas
- 13:21Building and integrating personas
- 18:18Future features and closing
related talks
17mYour realtime AI is ngmi
57mDesigning Human-AI Experience
23mExperimenting with Hardware and AI to Create Dynamic Live Streaming User Experiences
33mDesigning Voice Agents for Real Conversations
22mTop 10 AI tools for business in 2025
47mTesting the Agents That Look You in the Eye
17mTechnologies Enabling Real-Time Interaction With Digital-Humans
14mAI Beyond Knowing - Explore Emotion AI with Neura
