Virtual Avatar Engine with Real-time Motion Generation
RESEARCH / THESIS
Engineered an end-to-end Python pipeline for real-time generative animation.
Orchestrates LLMs, TTS, and MoMask motion synthesis concurrently via WebSockets.
Solved root motion drift using inertial blending heuristics for 60 FPS playback.
Click for live demo!
MoMask
WebSocket
VRM
Three.js
Asyncio
Ember AI
FOUNDING ENGINEER
Built the backend for a narrated memory experience startup.
Architected a fault-tolerant generation queue using Redis and BullMQ.
Optimized global delivery with Cloudflare R2, cutting egress costs by 50% while achieving <50ms audio-visual sync.
FastAPI
Redis
Voice Cloning
R2
Conversational Voice Agent with RAG
DAVIS INSTITUTE FOR AI
Designed a sub-second, full-duplex voice architecture.
Implemented a custom "barge-in" protocol handling raw mulaw streams, allowing users to naturally interrupt the AI without breaking state consistency.
Azure Speech
Vector Search
Audio Processing