Inspiration: Tackled the anxiety of public speaking by turning vague feedback into clear, actionable insights.
What it does: Captures timestamped notes during a live talk, sends them to a language model (via Hugging Face’s Inference API), and returns a performance summary, improvement tips, and issue flags—entirely in‑browser with no backend.
How you built it: Used React and Tailwind for a clean UI with a timeline view and one‑click analysis; integrated GPT‑2 directly from the frontend to avoid server infrastructure.
Key challenges: Prompt‑engineering GPT‑2 for structured output, dealing with CORS and client‑side API limits, and finding a backend‑free way to protect API keys.
Accomplishments: Delivered a fully client‑side feedback tool in under 24 hours with an intuitive UI and real‑time insights.
Lessons learned: Small prompt tweaks have huge impact, instruction‑tuned models outperform GPT‑2 for structure, and rapid pivoting on tech stacks pays off in hackathons.
What’s next: Swap in a more capable LLM (e.g. FLAN‑T5), add posture/voice analysis via webcam/microphone, enable video uploads, build a feedback dashboard, and introduce gamification/confidence scoring.
Log in or sign up for Devpost to join the conversation.