I noticed a pattern preparing for technical interviews. I could write perfect answers to questions. But saying them out loud, under pressure, felt completely different.
That's when I realised most interview prep tools train the wrong skill.
Typing and speaking are completely different skills
When you type an answer you can pause, edit, reorganise. When you speak you have to think linearly, stay coherent, and keep going even when you lose your train of thought. Most interview prep tools train the first skill. Real interviews test the second.
So I built DevPrep.app.
How it works
The AI acts as a technical interviewer. It asks you a question out loud. You speak your answer. It transcribes what you said, evaluates it, gives feedback, and asks a follow-up — just like a real interviewer probing deeper.
Not speech-to-text into a text box. A genuine two-way voice conversation.
The tech stack
ASP.NET Core 7 backend
React + TypeScript frontend
Azure OpenAI GPT-4o for question generation and evaluation
Web Speech API for desktop, Azure Cognitive Services for mobile
Azure Container Apps + Static Web Apps
Stripe for subscriptions
500+ flashcards included
Covering Azure, C#, React, TypeScript, Docker, Kubernetes, Python, SQL, Terraform, Git, Node.js and more. With progress tracking, review mode and session history.
What I learned building it
Getting GPT-4o to behave like a real interviewer was the hardest part. Early versions were too forgiving. Real interviewers push back and probe weak answers. Getting that behaviour consistently took a lot of prompt engineering.
Mobile was the second challenge — Web Speech API doesn't work reliably on mobile so I implemented Azure Cognitive Services Speech SDK for mobile users.
Try it
Free tier at devprep.app — no credit card required, works in your browser.
Would love feedback from anyone who's been through technical interviews recently.
Top comments (0)