DEV Community

Sushil
Sushil

Posted on

I Built an AI Voice Mock Interview Platform to Help Developers Prepare for Real Interviews

For the past few months, I've been building a side project called Voxa an AI-powered voice mock interview platform.

The idea wasn't born because I wanted to build another AI app.

It came from something much more personal.

The Problem

Like many developers, I've spent hours solving LeetCode problems, reading interview guides, and watching YouTube videos.

But when it came to actual interviews, everything changed.

I knew the concepts.

I had built projects.

Yet I often struggled to explain my thoughts clearly under pressure.

Most interview preparation websites only provide a list of questions. They don't simulate an actual conversation, ask follow-up questions, or explain where you went wrong.

And after a real interview, companies rarely tell you why you were rejected.

That made me wonder:

What if I could practice with an AI interviewer that actually talks to me?

The Idea

I wanted to build something that feels closer to a real interview than a quiz.

Instead of typing answers, users would speak naturally with an AI interviewer.

The interviewer should:

  • Ask technical or behavioral questions
  • Listen to the candidate's answers
  • Ask follow-up questions
  • Generate detailed feedback
  • Track progress across multiple interviews

That's how Voxa started.


Tech Stack

For this project I chose technologies that I'm comfortable building with while keeping the architecture scalable.

Frontend

  • React 19
  • Tailwind CSS
  • Clerk Authentication

Backend

  • FastAPI (Python)

Database

  • Supabase PostgreSQL

Voice

  • Vapi AI

AI Evaluation

  • OpenRouter LLM

Payments

  • Razorpay

Building the Interview Experience

The biggest challenge wasn't displaying a chat interface.

It was making the interview feel natural.

The application lets users choose:

  • Job role
  • Experience level
  • Interview duration

Once the interview starts, the AI conducts the conversation entirely through voice.

As the interview progresses:

  • The transcript is generated in real time.
  • The AI asks follow-up questions based on previous answers.
  • A timer keeps track of the session.
  • At the end, the transcript is sent for evaluation.

The goal was to make it feel less like answering predefined questions and more like talking to a real interviewer.


AI Feedback

After the interview finishes, Voxa generates a detailed report.

Instead of a single score, the report includes:

  • Overall interview score
  • Communication skills
  • Technical understanding
  • Confidence
  • Problem-solving ability
  • Strengths
  • Areas for improvement
  • Question-by-question evaluation
  • Personalized learning plan (Premium)

This was one of the hardest parts to design because AI responses can vary significantly.

Getting consistent, useful feedback required several prompt iterations and testing.


Challenges I Faced

Building an AI product is very different from building a traditional CRUD application.

Some of the challenges included:

  • Handling real-time voice conversations
  • Managing transcript synchronization
  • Designing useful AI prompts
  • Preventing failed AI responses from breaking the user experience
  • Managing subscription limits
  • Handling authentication across frontend and backend
  • Keeping API costs under control

I also spent a surprising amount of time improving small UX details like loading states, error handling, and progress indicators.


Lessons Learned

Working on Voxa taught me a few important lessons.

Build for a real problem

AI is exciting, but users only care if it solves something meaningful.

Keep the first version small

It's tempting to add every feature.

Shipping a working product is far more valuable than chasing perfection.

User experience matters

People notice smooth onboarding, good loading states, and thoughtful feedback much more than fancy animations.

Distribution is harder than development

Building the application took time.

Getting people to discover and use it is proving to be an even bigger challenge.


What's Next?

I'm continuing to improve Voxa based on user feedback.

Some ideas I'm exploring include:

  • Company-specific interview simulations
  • Resume-based interview generation
  • Coding interview support
  • Better analytics and progress tracking
  • Team and campus plans

Final Thoughts

Building Voxa has been one of the most rewarding projects I've worked on.

There are still plenty of things to improve, but seeing users complete mock interviews and receive actionable feedback makes the effort worthwhile.

If you're preparing for interviews, I'd genuinely love to hear what features would make a tool like this more useful for you.

Feedback is always welcome.

Thanks for reading!

Top comments (0)