I Built an AI Interview Coach That Runs Entirely in Your Browser (No API Keys Required!)
TL;DR: I created a free, privacy-first interview prep app that uses AI to generate questions and provide feedback—all running locally in your browser. No servers, no API keys, no data collection.
The Problem
Let's be honest: interview prep is expensive and stressful. Professional coaches charge $100+ per hour, and practicing alone doesn't give you the feedback you need to improve. I wanted to change that.
The Solution
I built an AI-powered interview coach that:
- 🆓 Is completely free (no hidden costs)
- 🔒 Runs 100% locally (your responses never leave your device)
- 🌐 Works offline (after initial setup)
- 🎯 Provides intelligent feedback on your answers
- ⚡ Requires zero configuration (no API keys!)
How It Works
The magic happens thanks to Transformers.js, a library that runs Hugging Face AI models directly in your browser using WebAssembly. Here's the flow:
- Choose a category (behavioral, technical, leadership, etc.)
- Answer AI-generated questions tailored to your category
- Get instant feedback on your responses
- Review your session summary with actionable insights
The first time you use it, the AI model (~250MB) downloads and caches in your browser. After that, everything is instant and works offline.
Why This Matters
Privacy First
Your interview responses are personal. With this app, they stay on your device. No cloud servers, no data collection, no privacy concerns.
Truly Free
No freemium tricks, no API limits, no credit card required. The AI runs on your computer, so there are no server costs to pass on to you.
Accessible to Everyone
Whether you're a student, career changer, or seasoned professional, you can practice interviews anytime, anywhere—even without internet after the initial setup.
The Tech Stack
- React + TypeScript for a robust, type-safe UI
- Transformers.js for browser-based AI
- flan-t5-small model for question generation and feedback
- Web Workers for non-blocking AI inference
- IndexedDB for model caching
Challenges I Overcame
Model Size vs Performance: Larger AI models give better responses but are too slow for browsers. After testing several options, I found the sweet spot with flan-t5-small (250MB).
Prompt Engineering: Getting consistent, useful feedback from the AI required careful prompt design and lots of iteration.
Browser Compatibility: Making WebAssembly and Web Workers work smoothly across different browsers took some creative problem-solving.
Try It Yourself
The app is open source and ready to use. Just clone the repo, run npm install && npm start, and you're ready to practice!
No setup, no configuration, no hassle. Just you and your AI interview coach.
What's Next?
I'm planning to add:
- Voice input/output for more realistic practice
- Progress tracking across sessions
- Custom question banks
- Multi-language support
Final Thoughts
Building this project taught me that AI doesn't have to be expensive or invasive. With modern browser technologies, we can create powerful, privacy-respecting tools that run entirely on the user's device.
If you're preparing for interviews, give it a try. If you're a developer, check out the code and see how browser-based AI works. And if you have ideas for improvements, contributions are always welcome!
Links:
- GitHub Repository
- Live Demo Built with React, TypeScript, and Transformers.js
Top comments (0)