Yara - Conversational AI Character

Disciplines

• Conversational AI
• 3D Character
• Narrative Systems
• Three.js

About this Project

Yara is an experimental conversational character based on a fictional protagonist from The YaraStar Trilogy, a series of science-fiction novels in which her identity and story originate.

Rather than treating AI as a text interface, the project asks what changes when a character can listen, respond, speak, and express emotion through a real-time 3D presence. A visitor’s speech is transcribed live, grounded in the story’s canon, interpreted through Yara’s personality, and returned through synthesised voice, facial expression, and synchronised lip movement.

Retrieval-augmented generation works with explicit story facts and character rules to preserve continuity. An emotion layer connects each response to vocal and facial behaviour, while speech visemes are synchronised to the audio clock and blended with ARKit-based expressions to make conversation feel continuous.

Through that combination of narrative identity, voice, memory, and expression, Yara explores how conversational AI can feel less like a utility and more like an encounter with a character.

Key Features

• Real-time voice-based conversation
• 3D character rendering with ARKit facial blendshapes
• Audio-synchronized lip movement using neural speech visemes
• Retrieval-augmented responses grounded in story canon
• Character-specific persona and continuity constraints
• Emotion-aware facial expression and voice behaviour
• Conversational memory across multi-turn interactions
• Live speech-to-text and synthesized voice responses
• Safety and moderation integrated into the character pipeline
• Responsive browser-based 3D interaction