We prototyped an AI-driven VR avatar demo for Meta Quest 3, aimed at elderly users who are far from their families. The goal: rapid development with realistic avatar that feels responsive, knowledgeable, and natural in conversation.

For speed, we used Convai with purchased assets, though a custom pipeline (OpenAI + TTS + STT via Azure/ElevenLabs) would offer more flexibility at the cost of extra development time.

Quick takeaways from the demo:

• Text generation and data exchange are good enough to sustain simple conversations.
• Built-in lipsync is basic — realistic results need phoneme-level data.
• Small animation details (eye movement, gestures) are critical for believability.
• Off-the-shelf solutions are great for rapid prototyping, but costly at scale.
• Higher realism requires more investment in modeling, animation, blendshapes, and optimization.

AI avatars are evolving fast — this demo shows how quickly a functional prototype can be built, and where the real effort lies when aiming for lifelike interaction.