Duolara – Your AI Voice Coach for Confident Interviews

Inspiration

Duolara was born out of a simple but powerful insight:

"Most students aren’t afraid of interviews because they lack knowledge — they’re afraid because they lack confidence."

During peer placement prep, we noticed even smart, well-prepared students struggled with basic HR questions like:

  • “Tell me about yourself”
  • “Why should we hire you?”

Not due to lack of answers, but due to anxiety and poor verbal delivery. Coaching was either expensive or unavailable — especially in Tier 2/3 cities. We decided to fix this using voice-based AI.


What We Learned

  • Voice interaction adds realistic pressure — great for practice.
  • Real-time feedback on tone, pace, and filler words builds confidence.
  • Students prefer private practice before speaking with real people.

How We Built It

  • Frontend: React Native + Vapi SDK for voice calls
  • Backend: Firebase (Auth + Firestore for transcripts & scores)
  • Voice AI: Transcripts processed by LLMs (Gemini/GPT) to generate structured feedback
  • Scoring System: Confidence is calculated based on tone, pitch, and pauses:

$$ \text{Confidence Score} = f(\text{Pitch Variability}, \text{Speaking Rate}, \text{Pause Density}, \text{Volume Stability}) $$

Where ( f ) is a weighted function based on behavioral communication patterns.


Challenges Faced

  • Creating a smooth, lag-free voice loop
  • Designing meaningful real-time scoring metrics
  • Making AI feedback feel like coaching — not just a transcript summary
  • Convincing shy users to speak to an AI voice agent

Future Scope

  • 🎯 Multilingual support (Hindi, Tamil, Bengali, etc.)
  • 🎯 Domain-specific tracks (tech, HR, case, design)
  • 🎯 Dashboards for college placement cells
  • 🎯 AI learning from anonymized practice data

Final Thought

“Rich kids hire coaches. We built one for everyone else.”

Duolara helps students speak their future into reality — one mock interview at a time.


Built With

Share this project:

Updates