Portfolio AI Companion
An English voice and text companion that follows visitors across the portfolio and answers from the page they are viewing.
- response language
- EN
- sign-ins required
- 0
- answer sources
- Linked
Project note
The first version of this portfolio assistant used AWS Bedrock to answer résumé questions behind an authenticated form. For the new site, the experience was reframed around a simpler product question: how can a visitor understand the most relevant evidence in a few minutes, without creating an account?
The current public mode uses one persistent LiveKit session for text and voice. It gives KIRA bounded context from the page a visitor is viewing, then uses Gemini and ElevenLabs to answer and speak. The floating companion starts with a lightweight 2D KIRA launcher. Its 3D avatar loads from the model CDN when a visitor first opens the conversation, then stays mounted when the panel closes. This keeps the model and rendering work out of an ordinary first visit. The engineering note on deferred avatar loading explains the state changes, browser checks, and remaining limits.
The interface is available as both a focused full-page experience and a floating panel throughout the site. Keyboard navigation, clear page highlights, mobile layouts, and reduced-motion preferences are treated as core interaction requirements rather than polish.