Inspiration At Penn State Behrend, our career services team does a fantastic job offering mock interviews to help students prepare for the workforce. However, there simply aren't enough staff members to accommodate the sheer volume of students who need practice. We realized that many candidates miss out on crucial, realistic interview preparation due to scheduling constraints and limited human resources. We built PrepprAI to bridge this gap, leveraging artificial intelligence to provide unlimited, on-demand, and high-quality interview practice for anyone, at any time.
What it does PrepprAI is an intelligent platform that generates hyper-realistic, AI-powered mock interviews tailored to a user's exact career goals.
Customized Context: Users can upload their resumes and specify their target company (like Google, Meta, or Amazon) and seniority level.
Dynamic Personas: The app dynamically constructs a high-stakes environment mirroring that company's culture. Users can practice against diverse AI personas, ranging from "The Stress Tester" to "The Chill Mentor," to prepare for any interview style.
Real-Time Emotion Intelligence: Using advanced facial tracking, PrepprAI decodes micro-expressions (confidence, hesitation, stress) during the interview.
Actionable Feedback: Alongside sentiment analysis of verbal responses, users receive specific, timestamped insights to refine their narrative and physical presence before the real thing.
How we built it We built the frontend of our application using React.js and Tailwind CSS to create a clean, responsive, and highly customizable user interface. Our backend architecture is driven by Python, utilizing MongoDB for flexible and scalable data storage.
To power the core intelligence of the app, we integrated two main technologies:
Gemini AI: Used for deep contextual understanding, generating dynamic, company-specific questions, and providing detailed transcript analysis.
OpenFace: Implemented to detect 68 facial landmarks, allowing us to track micro-expressions and provide real-time emotional and stress analysis during the video feed.
Challenges we ran into Our biggest hurdle was scope management. We had a massive list of ambitious ideas, and figuring out how to merge our different concepts and integrate complex technologies within a strict 24-hour timeframe was incredibly difficult. Getting the OpenFace computer vision models to communicate smoothly with the web frontend and the Gemini backend in real-time presented significant technical roadblocks that required a lot of rapid problem-solving.
Accomplishments that we're proud of We are incredibly proud of how much we learned and implemented in such a short period. Prior to this hackathon, our team had never used Docker, React, or built comprehensive web-based applications from scratch. Stepping completely out of our comfort zones to build a functional, full-stack web app with complex AI and computer vision integrations is a massive win for us.
What we learned We learned the vital importance of MVP (Minimum Viable Product) thinking—how to prioritize core functionalities over "nice-to-have" features when racing against the clock. Technically, we gained invaluable, hands-on experience with modern frontend frameworks (React, Tailwind), containerization (Docker), and the intricacies of connecting advanced machine learning models to live web applications.
What's next for PrepprAI We want to take PrepprAI to the next level by fully transitioning our infrastructure to the cloud. Our next major step is adding cloud integration via AWS for more robust video processing, scalable database hosting, and seamless continuous deployment. This will allow PrepprAI to handle a larger volume of users practicing their interview skills concurrently.
Log in or sign up for Devpost to join the conversation.