Inspiration
Presenting is a skill, but most people only get meaningful feedback when it actually matters — during a class presentation, interview, pitch, or meeting. I wanted to make that feedback available before the stakes are high.
presentIQ was inspired by the idea of giving people a private coach they can practise with repeatedly: one that can listen to their speech, observe their delivery, and point out things they might not notice themselves.
I also wanted to approach presenting from an inclusive education perspective. Some people struggle with public speaking, confidence, body language, or finding the right words. I wanted to build something that gives them a private, judgement-free environment to practise and improve at their own pace.
What it does
presentIQ is an AI presentation coach for people who struggle with presenting.
Users first take a short quiz to test their understanding of their material, then record their presentation. presentIQ analyses both what they say and how they deliver it, looking at speech, word choice, gestures, body language, and overall delivery.
Instead of simply saying “good job,” presentIQ turns the analysis into actionable feedback — highlighting strengths and showing the presenter where they can improve.
The core experience is:
Quiz → Record → Analyse → Feedback → Improve
How I built it
I built presentIQ as a web application with a React/Vite frontend and a Python/FastAPI backend.
The presentation analysis pipeline combines multiple forms of analysis:
- OpenAI Whisper for speech transcription and analysis
- Computer vision for observing presentation behaviour
- Gesture analysis to evaluate body language and movement
- Language analysis to assess wording and word quality
- Structured feedback that turns the analysis into practical recommendations
The goal was to combine these signals rather than judge a presentation from speech alone.
The repository is structured into separate frontend and backend projects, making it possible to run the complete application locally.
Deployment
For the hackathon submission, I wasn’t able to host the backend publicly because the backend requires server-side processing resources, and the free hosting options I evaluated either had usage limitations or would expire after a limited period.
Rather than present a broken or incomplete live deployment, I hosted a web preview of the frontend so judges can immediately see and interact with the interface.
The full application, including the backend and analysis pipeline, can be previewed by running the project locally using the setup instructions in the repository.
This means the hosted preview demonstrates the user experience, while the GitHub repository contains the complete project for local execution.
Challenges I ran into
The hardest challenge was turning something as subjective as “good presenting” into measurable signals.
Speech, body language, gestures, pacing, and word choice all tell different parts of the story. I had to think carefully about how to analyse these independently and then combine them without overwhelming the user with meaningless scores.
I also had to deal with the practical challenges of processing recorded audio and video, connecting multiple analysis systems, and presenting the results in a way that actually helps someone improve rather than simply giving them a wall of metrics.
Another challenge was deployment. Running the complete analysis pipeline requires backend resources that weren’t practical to keep publicly hosted using the free deployment options available to me during the hackathon. I therefore focused on making the project fully runnable locally while providing a hosted frontend preview for judges.
Accomplishments that I’m proud of
I’m proud that I built a working concept around a problem that is often treated as purely human rather than something technology can meaningfully assist with.
Most importantly, presentIQ doesn’t just analyse a transcript. It attempts to understand the whole presentation experience — what you said, how you said it, and how you physically delivered it.
I’m also proud of building it with inclusive education in mind: giving people a private, judgement-free environment where they can practise as many times as they need.
I also successfully brought multiple technologies together into a single workflow — speech recognition, computer vision, language analysis, and structured feedback — rather than building a simple text-based AI wrapper.
What I learned
I learned that building an AI product isn’t just about connecting models together.
The difficult part is deciding what signals actually matter, how reliable those signals are, and how to turn raw model output into feedback a real person can understand and act on.
I also learned that multimodal AI becomes much more useful when different inputs — speech, vision, and language — are combined around a specific user problem.
Most importantly, I learned to make pragmatic engineering decisions under time and infrastructure constraints. When a production deployment wasn’t financially practical during the hackathon, I made sure the complete application remained runnable locally instead of sacrificing the core functionality.
What’s next for presentIQ
Next, I want to make presentIQ much more precise and personalised.
I’m looking at:
- More detailed body-language and gesture analysis
- Better presentation scoring
- Improved speech metrics such as pacing, pauses, and filler-word detection
- More advanced word-choice and clarity analysis
- Personalised recommendations based on previous presentations
- Progress tracking across multiple practice sessions
- More robust backend infrastructure for a publicly hosted version
Ultimately, I want presentIQ to become a personal presentation training loop:
Practise → Analyse → Receive Feedback → Practise Again
The goal is to help users turn presentation anxiety and uncertainty into measurable improvement over time.
Log in or sign up for Devpost to join the conversation.