Inspiration

I’ve always found public speaking a bit hard and nerve-wracking. I wanted to build something that would allow you to practice with a reliable teacher that wont hold back with its critiques to make you improve at the fastest rate and will help you gain confidence before the actual presentation.

What it does

It’s a dual-mode coach funtioning according to your needs, Live Practice listens to you in real-time and whispers simple tips if you're too fast or repetitive. Video Review lets you upload a recording so the AI can watch your body language and visuals, giving you a comprehensive critique.

How I built it

I used a Node.js backend and a clean Vanilla JS frontend. The "brain" is powered by Gemini 2.5 Flash for multimodal video analysis and their Preview TTS to give the AI a high-quality, natural human voice.

Challenges I ran into

The biggest headache was the time where the camera and speech engine kept fighting for control which was a problem that kept me up for a few days. I also spent a lot of time debugging API quotas and experimental model names to keep everything running smoothly.

Accomplishments that i am proud of

Getting the Gemini API to actually work, and when i tested the prompt i have used for the api to analyze you and seeing it transition from just a transcript to a coach that can talk back to you was a huge like yay moment.

What I learned

I learned that feedback needs to be sparse to be effective, sometimes the best coaching is staying silent. I also got a crash course in audio headers and handling real-time data streams. and i also discovered about anti gravity :)

What's next for GeminiAI Presentation Critic

I want to add slide deck analysis to see if your visuals match your words, and i guess id like to have the ai evaluate both your view (movements) and speech in the same time in the live presentation one.

Built With

Share this project:

Updates