I need a working prototype of video-based chat rooms where every participant appears as an AI-driven avatar. The visual side should come through HeyGen (or a comparable engine) so the avatar can mirror basic facial expressions while speaking. Audio must be lip-synced and the avatar’s voice routed through an OpenAI model that answers real-time user queries—ChatGPT provides the intelligence layer, not just canned lines. Core flow • A user joins a room through the browser, camera on. • Their live video is replaced by the animated avatar. • Anything they say or type is sent to ChatGPT; the response is spoken back by the avatar in natural voice, with matching mouth and facial cues. Technical expectations • WebRTC or a similar stack for low-latency video chat. • HeyGen (current preference) integrated for avatar rendering, with the hooks exposed so I can later swap to my own pipeline via a clean REST/GraphQL API. • OpenAI GPT-5 (or the latest available) handling conversational logic specifically for user queries; moderation and other tasks are out of scope for now. • Source code, clear setup instructions, and an abstraction layer that keeps the avatar engine and AI service pluggable. - More, will be disscuted. If you have shipped something similar, especially with WebRTC, HeyGen, or direct OpenAI integrations, that experience will help accelerate the build. I’m ready to iterate quickly and test early builds, so well-structured, modular code is essential. Also, i can present and give access at an platform wich have much of the desired capability but unfortunately don't work very good, for understand better my request.