Personal project · Full-stack SaaS
Mockly — AI Mock Interview Platform
Job seekers lacked realistic interview practice with instant feedback. Mockly generates role-specific questions with GPT-4o-mini, transcribes spoken answers in real time and scores them on clarity, depth, structure and relevance.
Problem
Practising interviews alone gives no feedback, and practice sessions with other people are hard to schedule. I wanted a tool that asks realistic questions for a chosen focus (frontend, backend, system design or behavioral), listens to a spoken answer and explains right away what to improve.
Solution
- React + TypeScript single-page app (Vite, Tailwind) talking to a Node.js/Express API over REST and WebSockets.
- Adaptive question generation with GPT-4o-mini across four interview categories, without repeats.
- Real-time speech-to-text by streaming audio chunks over a WebSocket to Deepgram, with automatic fallback to batch transcription with OpenAI Whisper.
- Answer analysis with 0–100 scores for clarity, depth, structure and relevance, plus filler-word detection in the transcript.
- Questions and feedback read aloud with OpenAI TTS or ElevenLabs voices.
- Supabase (PostgreSQL with Row-Level Security) for the waitlist and beta feedback, Resend for transactional email, and Zod validation on every input.
- A mock mode with deterministic responses for every AI endpoint, so the whole app runs locally without API keys.
Tech stack
- React
- TypeScript
- Node.js
- WebSockets
- GPT-4o-mini
- Whisper
- Deepgram
- ElevenLabs
- Supabase
Results
- Live demo deployed, with the frontend on Vercel and the API on Render.
- Four external AI providers behind one API, each with a fallback path.
- Waitlist and beta-feedback flows in place to collect input from early users.