🚀 SKILLGPS – AI-Powered Interview Coach

💡 Inspiration

Most people don’t fail interviews because they lack knowledge —
they fail because they don’t know what they’re weak at.

Existing platforms provide generic questions and surface-level feedback, but none focus on targeted improvement.

We built SKILLGPS to solve this:

An AI system that identifies your weak skills and trains you like a real interviewer.


🧠 What it does

SKILLGPS is a personalized AI interview coach that:

  • 🔍 Detects skill gaps from user profiles
  • 🎯 Generates questions only from weak areas
  • 🎤 Conducts live AI mock interviews (voice + text)
  • ⚡ Adapts difficulty in real-time based on answers
  • 📊 Provides strict, realistic evaluation (Correct / Partial / Incorrect)
  • 📈 Tracks skill-wise improvement over time

👉 The result: Users don’t just practice — they improve faster and smarter.


🛠️ How we built it

  • Backend: FastAPI (Python)

    • Adaptive interview engine
    • Skill-based question generation
    • Real-time evaluation + feedback system
  • Frontend: HTML, CSS, JavaScript

    • Modern dark UI with glassmorphism
    • Voice interaction using Web Speech API (STT + TTS)
    • Live interview experience with animations and feedback
  • Core Innovation:

    • Skill-gap driven questioning
    • Adaptive interview flow
    • Honest evaluation (no fake praise)

⚙️ Challenges

  • Building a realistic evaluation system (avoiding fake positivity)
  • Designing adaptive interview logic based on answer quality
  • Handling smooth voice interaction (STT + TTS)
  • Creating a UI that feels like a real product, not a prototype

📚 What we learned

  • How to design adaptive AI systems, not static tools
  • Importance of accuracy over politeness in feedback
  • Building real-time AI experiences with voice
  • Thinking like a product builder, not just a developer

🏁 Final Impact

SKILLGPS transforms interview preparation from: ❌ Random practice
➡️
Targeted, AI-driven improvement

It doesn’t just ask questions —
it trains you to get better.

Built With

Share this project:

Updates