Professor KIA: Voice-First AI Tutoring Platform
- Built a voice-first AI tutoring platform where students speak to Professor KIA over a shared freehand whiteboard by integrating Claude API (LLM + vision), Deepgram STT, and ElevenLabs TTS through a single WebSocket in a FastAPI / Next.js 14 full-stack architecture
- Engineered a handwriting synthesis pipeline using fonttools to extract B´ezier curves from TTF glyph outlines, converting them into time-sequenced stroke points rendered as animated writing on a tldraw canvas with automatic position rebasing
- Implemented real-time barge-in detection with asyncio and Web Audio API PCM scheduling, enabling sub-200ms interruptions and a proactive board-analysis loop that corrects student mistakes after 4s of silence

