Intervue

• 

  System Architectural Flow

Inspiration 💡

The Problem Job interviews are the scariest part of being a student. We found that:

• 93% of job seekers feel massive anxiety before an interview (huntr).
• 1 in 4 hiring managers say college grads aren't ready for the real world (Intelligent).
• Most practice tools are just boring lists of questions. They don't feel real, and they don't give you actual feedback on how you look or sound.

The Solution: ** We built **Intervue because 3 out of 4 of us on this team just finished internship interviews and realized we weren't ready for the "vibe" of the room. We wanted to create a simulator that feels 100% real. It’s not just a chatbot it’s an AI that sees you, hears you, and challenges you like a real recruiter from a top company.

What it does 🚀

Intervue is the ultimate mock interview simulator.

• Smart Behavioral Interviews: Our AI doesn't follow a script. It actually listens to you and asks follow-up questions based on what you said.
• FAANG-Style Technical Lab: We built a coding environment that looks and feels exactly like the ones used by the world’s biggest tech companies.
• 3-Model Multimodal Feedback: We use three different Gemini Pro models to watch your facial expressions, listen to your tone, and check your code all at the same time.
• Video Review: After you're done, you get a report card with timestamps. Click a mistake, and it takes you right to the video of you making that mistake so you can fix it.

How we built it 🛠️

We used a high-performance stack to make sure the app felt smooth and professional:

• Frontend: React and CSS for a sharp, "insane" UI.
• Backend: Flask to handle all the heavy data moving between the user and the AI.
• Local Speech-to-Text: We ran OpenAI Whisper locally on our machines to convert user speech to text instantly, ensuring low latency and better privacy.
• The AI Brain: 3 Google Gemini Pro models working together. One for video, one for audio, and one for code logic.
• The Voice: ElevenLabs to give our AI a human-sounding voice that changes based on how hard you want the interview to be.

Challenges we ran into 🧗

The biggest fight was getting the frontend and backend to talk to each other. Our frontend team was moving so fast that by the time the backend was ready, the UI was already on a whole different level.

We also struggled to sync the audio and video feedback. We wanted to make sure that if the AI gave you a tip about a specific answer, the timestamp would take you exactly to that spot in the video. It took a lot of debugging to get those clocks to line up perfectly

Built With

• css
• elevenlabs
• flask
• google-gemini
• javascript
• python
• react
• whiser