Inspiration
As a professional who’s fumbled through high-stakes pitches and watched peers struggle with public speaking, I saw a gap: personalized, instant feedback was only accessible to those who could afford expensive coaches. I wanted to build something that democratized that expertise—using AI to make pro-level analysis available to everyone, from students to startup founders.
What it does
PublicSpeaking Coach lets users record/upload speech videos, then AI breaks down verbal (speaking rate, filler words), non-verbal (eye contact, posture), and content (structure, clarity) metrics. It delivers a score, detailed feedback, and actionable tips—plus tools to practice, track progress over time, and compare videos to see growth.
How we built it
We started with core tech stack: React for the frontend (intuitive UI for video upload/feedback), Node.js/Express for backend logic. AWS hosted video processing; FFmpeg handled video trimming. OpenCV was key for computer vision (non-verbal analysis), and Stripe API integrated subscription tiers (Free/Pro/Team). We built modular features—recording module, feedback engine, progress graphs (Chart.js)—and tested rigorously with real users (professionals, students) to refine UX.
Challenges we ran into
The biggest hurdle was accurate non-verbal analysis: eye contact tracking required fine-tuning OpenCV to account for different lighting/camera angles, and gesture "naturalness" scoring needed training AI on diverse datasets. We also struggled with video processing speed—initially, analysis took 2+ minutes, so we optimized FFmpeg workflows and cloud resource allocation to hit 30-60 second processing. Subscription billing edge cases (downgrades mid-cycle, payment failures) also required tight logic to avoid user friction.
Accomplishments that we're proud of
We built a fully functional platform that delivers precise, actionable feedback—users report 20%+ improvement in filler word reduction and eye contact within a month. The responsive design works seamlessly across devices, and Pro/Team tier users love the PDF report and team management features. Most importantly, we hit our goal: making high-quality public speaking feedback accessible to all tiers of users.
What we learned
I deepened my understanding of AI/computer vision integration with web apps, and learned to balance technical precision with user-centric design (e.g., simplifying feedback language so it’s actionable, not just technical). I also gained insight into subscription business models—how to structure tiers that feel fair and provide real value at every level.
What's next for this project
First, expand practice tools: add AI-powered speech generation to let users draft content, and integrate personalized practice plans based on their feedback gaps. We’ll also build mobile web optimizations (beyond responsive design) and explore integrations with video conferencing tools. Long-term, we want to add multi-language support and team collaboration features (e.g., shared feedback for corporate teams).
Log in or sign up for Devpost to join the conversation.