Sugar Rush
Team led by an MSCS graduate from Stevens and Applied AI Engineer skilled in real-time voice APIs, agentic architectures, and model fine-tuning.
YouTube Video
Project Description
Aria is an AI voice assistant embedded directly into a portfolio website that completely replaces the traditional chat widget paradigm.
Instead of a text box, visitors interact through an audio-reactive UI — a 5-layer visual element with glass morphism effects that pulses,
glows, and changes color in real-time based on conversation state (listening, speaking, connecting, error). The entire viewport responds with
a cinematic edge glow effect that turns the browser into an immersive voice session.
What makes the interface the innovation: There is no chat window. No text input. No message bubbles. The UI is the conversation — a pulsing
orb, a glowing screen edge, and a floating hint pill that teaches users naturally. When scrolled past the hero, a floating mini-orb persists
at the bottom-right with Cmd+K keyboard shortcut access, keeping the assistant one keypress away at all times.
Claude-powered reasoning: Aria uses Claude as her brain — not just for Q&A but for real-time visitor qualification. When Claude detects
recruiter/hiring manager signals (company mentions, hiring language, talent acquisition phrases), it naturally steers the conversation toward
scheduling a meeting. It checks Google Calendar availability via tool-calling, offers open 30-minute slots, collects attendee details
conversationally, and creates calendar events with Google Meet links — all through voice, zero clicks. Claude’s reasoning drives the entire
decision tree: when to qualify, when to schedule, when to pivot topics, and how to match the visitor’s energy level.
Tech stack: Next.js 15 (App Router), React 19, TypeScript, Claude (Anthropic) for reasoning and conversation, Vapi.ai (voice SDK + WebRTC),
Framer Motion (compositor-only animations at ~15fps throttled volume updates), ElevenLabs (voice synthesis), Google Calendar API (OAuth2 +
freebusy scheduling), Tailwind CSS v4, Vercel. Claude Code was also used as the development environment for rapid prototyping throughout the
hackathon.
Key UI innovations:
- Audio-reactive orb with 5 rendering layers (outer glow, pulsing ring, core sphere, glass highlight, inner shimmer)
- Full-viewport EdgeGlow effect using GPU-optimized inset box-shadows (opacity-only, will-change: opacity, contain: strict)
- Auto-hiding VoiceActivationHint pill with platform-specific shortcut detection (Mac vs Windows)
- prefers-reduced-motion accessibility support throughout
- Portal-rendered overlays for proper z-stacking without layout interference
Prior Work
Major components of the the portfolio were built early on. Voicebot was built at this hackathon