Lumina Presenter

Main UI
HUD Capture

Inspiration

Public speaking anxiety affects millions of people and can trigger visible stress responses such as fidgeting, eye contact avoidance, and cognitive stalls. These signals often hurt confidence and performance, especially in high-stakes settings like interviews and pitches.

We wanted to build a tool that acts as a real-time support system, helping people manage presentation anxiety using only their laptop camera

What it does

Lumina-Presenter turns a single webcam into a real-time presentation coach. • Uses MediaPipe to track body and face landmarks • Detects fidgeting via temporal movement variance • Estimates gaze direction using MediaPipe Face Mesh • Displays a transparent, always-on-top HUD built with PyQt6 • Triggers live alerts for gaze loss, instability, and cognitive stalls

It acts as a “second brain” during pitches and interviews.

How we built it

• Computer Vision Layer: Extracted skeletal + facial landmarks from webcam feed
• Signal Processing Layer: Used NumPy/SciPy to calculate variance, thresholds, and filtered signals
• HUD Frontend: Built a frameless transparent overlay with smooth animations
• Calibration System: Personalized baseline posture, movement, and gaze thresholds

Challenges we ran into

Filtering natural movement (breathing/blinking) without false alerts • Estimating accurate gaze from a single monocular camera • Building a click-through transparent overlay without breaking usability Getting speech recognition to properly work to identify whether a user is on track

Accomplishments that we're proud of

Built a full biometric feedback system using only a laptop webcam • Achieved real-time gaze + movement inference • Designed a professional-grade HUD that feels production-ready • Created adaptive calibration instead of static thresholds

What we learned

Signal smoothing is critical in real-time AI systems

Clear team architecture (Vision → Signal → HUD) speeds development

What's next for Lumina Presenter

Allow users to save recorded presentations
• Generate post-session analytics:
• Create progress tracking across multiple sessions
• Add voice pacing and filler-word analysis