Aura: AI Public Speaking Coach

By The HackStreet Boys | HackPSU 2025 Submission

Inspiration

Our idea for Aura came from personal experience. Two of our team members struggle with public speaking anxiety, and one of us works as a Teaching Assistant for CAS 100A, a course focused on speech and presentation skills. We saw first-hand how difficult it can be for students to gain confidence, receive personalised feedback, and track their progress. This inspired us to build a digital coach that can help anyone improve their public speaking skills in a comfortable and affordable way.

What It Does

Aura is an AI-powered public speaking coach that analyses a user’s voice, posture, gestures, and expressions in real time. It provides immediate feedback and generates a detailed performance report after each session. The goal is to help users build confidence and improve delivery through objective, data-based insights.

Core Features

Voice and Speech Analysis Detects filler words, pace, and clarity Provides vocal confidence scores and improvement advice Checks topic adherence using keyword-based analysis Body Language Detection Uses MediaPipe Pose and FaceMesh for real-time tracking Analyses posture, eye contact, gestures, and smile frequency AI-Powered Coaching

Uses the Gemini API for summarised feedback Generates timestamped insights such as “slouched at 0:25” or “good gesture at 1:12” Calculates an overall confidence and presence score Visual Reports

Session dashboard showing confidence trends, topic adherence, and vocal metrics Exportable reports with strengths and improvement areas

How We Built It Frontend: React, TypeScript, Vite, Tailwind CSS Vision AI: MediaPipe Pose and FaceMesh Audio Analysis: Web Speech API with Gemini summarisation Backend (optional): FastAPI for server-side processing Hosting: Vercel with a custom .tech domain

Challenges Synchronising voice and body analysis in real time Fine-tuning gesture detection to avoid false positives Designing an intuitive dashboard for user feedback

Accomplishments Built a working AI-based coach fully in the browser Combined computer vision and audio analysis successfully Developed an interface that encourages continuous improvement

Future Plans Support for multiple languages and accents Emotion and engagement detection Gamified practice sessions and progress tracking

Share this project:

Updates