InterPrep

Inspiration

In a world where virtual interviews are the norm, there should be a tool to help you master them. These interviews can be nerve wracking and your friends only have so much time to help you with mock interviews. On top of that, it's often hard to get precise, real-time, and personalized feedback on two things that really matter in an interview: eye contact and expressions.

What it does

InterPrep is an AI companion that helps you get better at the skill of interviewing. Start off by selecting from our list of common interview questions (or select one using the 'Random Question' button!). Once you're ready, hit 'Record' and provide your answer as you would in a real interview. After completing your answer, hit 'Stop & Analyze' to get instant insights on your response. Get an eye contact and expression score along with feedback on where you can improve.

Repeat with the same question or choose a different question to keep practicing. For each recording within this session, you will be able to view your history on the top right, giving you the motivation to keep improving.

How does InterPrep Relate to Cognitive Wellness?

Interviews are naturally stressful and our aim was to help people walk into them feeling more confident. By getting personalized feedback, users can improve their interview skills and ultimately feel more confident in their skills

How we built it

We built this using the open-source React framework, Next.js. We used Next.js for a many reasons with the main being its ease of use to build full-stack, production-ready web applications, support for server side rendering which was useful for integration with Google's Mediapipe Face Land Marker API, as well as its script optimization and CSS support.

Challenges we ran into

We ran into the challenge of finding a lightweight yet powerful facial tracking library that accepts live video feed as input. Since none of us had experience with computer vision tasks before this took initial trial and error. Ultimately, we landed on Google's Mediapipe Facelandmarker task which uses the Face mesh model to locate 478, 3-dimensional landmarks on detected faces. This API provided plug and play code for the web platform.

Accomplishments that we're proud of

We're proud of utilizing and learning Next.js in a short 24 hour time frame. Along with that getting familiar with how to use off the shelf models such as Facelandmarker was a great

What we learned

We learned how to utilize Next.js and smaller libraries like Chart.js for additional features. Learning how to use off the shelf models such as Facelandmarker was experience in how to quickly utilize pre-existing open source models to accomplish specific inferencing tasks.

What's next for InterPrep

InterPrep can improve in a number of way, some ideas are:

• Countdown timer before the recording actually starts
• Save video recording, allowing the user to rewatch their previous trial
• User login/signup so that session history can be saved as user history instead
• Larger question bank and more personalized, AI driven feedback

Built With

• chart.js
• css
• google-mediapipe-tasks
• javascript
• mediapipe-facelandmarker-api
• next.js
• node.js