2025/06 - NoteFlow
- Architected a full-stack Canva extension using React, TypeScript, and Flask to generate AI-driven speaker notes from slide input in real time that integrated Hugging Face’s TinyLlama model for on-demand text generation.
- Designed a modular prompt injection system supporting dynamic tone, pacing, and structure inputs to enable 5 user-selectable speaking tones via system prompt variations.
- Implemented pacing constraints using token estimation and post-generation WPM analysis to align output with user-defined durations (15s–5min).
- Developed frontend features including live tone previews, duration sliders, and toggleable add-ons (Main Idea, Transitions), using Canva’s App UI Kit to optimize for responsive performance with sub-10s round trip latency.
You can view the source code on GitHub and the demo on YouTube.