HIREPREP-AI

HirePrep AI: Your personal AI career coach for interactive mock interviews and tailored study roadmaps.

Comment

Inspiration

The job market is increasingly competitive, and many candidates struggle to bridge the gap between their current resume and the specific requirements of their target roles. While traditional interview prep tools offer generic questions, they lack the personalization, dynamic adaptability, and real-time feedback that an actual interview provides. We were inspired to democratize access to high-quality, personalized career coaching. By leveraging Large Language Models (LLMs) and real-time voice synthesis, we envisioned a platform that could act as a 24/7 personal career coach—one that doesn't just give you a list of questions, but actively listens, coaches you in real-time, and builds a customized roadmap for your success.

What it does

HirePrep AI is a full-stack, AI-powered career platform that transforms your resume into an actionable strategy.

  • Deep Resume Analysis: It maps your career history against specific job descriptions to identify strong suits and critical skill gaps.
  • Instant Roadmaps: It generates custom, day-by-day preparation plans to help you bridge those gaps.
  • Live Voice Interviews: Candidates can engage in real-time, voice-enabled mock interviews that feel like talking to a real recruiter.
  • Real-time Coaching: The platform acts as a friendly interview coach, providing instant, per-answer feedback during live sessions.
  • ATS-Optimized Resumes: It automatically rewrites and generates tailored, ATS-friendly PDF resumes specific to your target role.
  • AI Job Matcher: It cross-references your profile with live job boards (via JSearch) to find the perfect opportunities.

How we built it

We built HirePrep AI using a decoupled monorepo architecture for optimal scalability:

  • Frontend: Built with React 19 and Vite, featuring a modern glassmorphism design system styled with SCSS. We heavily utilized the Web Speech API for real-time speech-to-text (STT) and text-to-speech (TTS).
  • Backend: A robust Node.js and Express server, utilizing MongoDB for data persistence.
  • AI Engine: The core intelligence is driven by Google Gemini (gemini-2.5-flash). We enforced strict Zod schema validations to ensure the AI consistently returns structured JSON data.
  • Performance: We integrated Redis as a caching layer to store expensive AI responses (slashing latency and API costs) and to handle rate-limiting.
  • PDF Generation: Puppeteer is used on the backend to dynamically render AI-generated HTML into downloadable PDFs.
  • Deployment: The frontend is hosted on Vercel, while the backend is fully containerized with Docker to seamlessly manage complex system dependencies.

Challenges we ran into

Building a real-time, AI-driven application brought several significant technical hurdles:

  1. Structured AI Outputs: Getting an LLM to consistently return complex, nested JSON objects (like a full interview report) without hallucinations or malformed syntax was difficult. We solved this by using strict prompt constraints and Zod schemas to validate outputs on the server.
  2. Containerizing Puppeteer: Running headless Chromium inside a Docker container is notoriously tricky. We had to carefully configure our Alpine Linux base image to include native font rendering libraries while managing memory limits to prevent crashes during PDF generation.
  3. Real-Time Voice Latency: Creating a natural conversational flow meant we had to aggressively minimize the delay between the user speaking and the AI responding. Synchronizing the frontend Web Speech API state with backend AI generation required precise asynchronous state management.

Accomplishments that we're proud of

  • Seamless Voice Interaction: We are incredibly proud of the live mock interview feature. Getting the speech recognition, AI processing, and voice synthesis to work together smoothly creates a genuinely immersive experience.
  • Automated ATS PDF Generation: Successfully generating fully formatted, ATS-compliant PDF resumes dynamically on the server side via Puppeteer.
  • Robust Architecture: Setting up a clean, scalable monorepo with a fully containerized backend and Redis caching layer ensures the platform is production-ready and highly performant.

What we learned

Building HirePrep AI was a massive learning experience across the entire stack. We deeply enhanced our understanding of prompt engineering and how to force LLMs into strict structural constraints. We mastered the DevOps intricacies of Dockerizing heavy Node.js applications that rely on external native binaries. Additionally, we learned how to effectively leverage Redis for more than just simple storage—using it for session blacklisting, rate limiting, and caching expensive external API calls.

What's next for HIREPREP-AI

We have big plans for the future of HirePrep AI!

  • Video & Behavioral Analysis: Expanding beyond voice to analyze facial expressions, eye contact, and body language during mock interviews.
  • Direct Application Integrations: Allowing users to apply to the jobs found via our AI Job Matcher directly from the platform.
  • Peer-to-Peer Mock Interviews: Introducing a collaborative mode where candidates can practice with peers in a supervised AI environment, receiving automated feedback on both the interviewer's and interviewee's performance.

Built With

Share this project:

Updates