YumPrep

Inspiration

The inspiration for YumPrep came from my personal, often brutal, job-hunting experiences. I realized that no matter how many hours I spent preparing, the high-stakes environment of a real interview could still make my mind go completely blank when faced with a "curveball" question. I felt there was a gap between traditional mock practice and the actual "battlefield." I wanted to build a "safety net"—a smart co-pilot that provides a tiny nudge of confidence exactly when you freeze.

What it does

YumPrep is an all-in-one AI interview co-pilot and training ground:

Real-time Live Coaching: It securely processes ongoing interview audio and delivers instant, concise bullet-point hints to your screen.

Daily Practice: Short, bite-sized daily questions to build muscle memory and keep communication skills sharp.

The "Offer Challenge": A high-pressure simulation mode designed to mimic intense final-round interviews.

Deep Analysis: Provides a multi-dimensional scorecard and video-based confidence charts to highlight blind spots.

How I built it

We built the core engine using the Gemini Live API for low-latency audio processing and real-time hint generation. The frontend is developed with React Native to ensure a seamless cross-platform experience. For the video analysis pipeline, we integrated NVIDIA NIM models alongside Gemini to track facial expressions and body language, generating a frame-by-frame confidence metric.

Challenges I ran into

One of the biggest hurdles was the latency of real-time delivery. For a hint to feel natural, it needs to appear almost instantly without interrupting the user's conversational flow. We spent a significant amount of time optimizing our prompt engineering and API streaming to ensure the "co-pilot" feels like a natural extension of the user's thoughts.

Accomplishments that I'm proud of

We are incredibly proud of the Real-time Live Coaching feature. Seeing the AI accurately pick up on a complex technical question and provide a relevant hint in under a second felt like magic. We also successfully created an AI interviewer that doesn't just read scripts but actively "pushes" the candidate based on their previous answers.

What I learned

Building YumPrep taught us a lot about the nuances of human communication. We learned that "what" you say is just as important as "how" you say it, which led us to prioritize the non-verbal analysis (tone and body language) as much as the content of the answers.

What's next for YumPrep

We plan to expand our library of specialized mock interview prompts for niche industries like high-frequency trading and specialized medical roles. We are also looking into integrating more personalized feedback based on the specific job descriptions users are applying for.

Built With

• react