IntervYou - AI Mock Interview Platform

IntervYou is a voice-first AI mock interview platform that conducts real, adaptive conversations — not scripted quizzes. The AI interviewer (Aria) listens to your answers, follows up on vague points, pushes back when answers are weak, and acknowledges strong responses before moving on. After each session, you get a structured feedback report with scores, strengths, and areas for improvement.

Architecture

Next.js Client (App Router) ←→ Express API ←→ Postgres (Neon)
       ↕                              ↕
  Vapi (WebRTC voice)           Groq (LLM feedback)
       ↕
  Groq (live conversation via Vapi)

• Frontend: Next.js 16, JavaScript, Tailwind CSS v4 - dark-mode glassmorphism UI
• Backend: Node.js + Express - REST API with JWT authentication
• Database: PostgreSQL on Neon (serverless) - users, sessions, feedback reports
• Voice AI: Vapi manages the full voice pipeline (STT → LLM → TTS) via browser WebRTC. The assistant is built as a "transient assistant" per session , no dashboard config needed.
• LLM: Groq (openai/gpt-oss-120b by default, override with GROQ_MODEL) — powers both the live conversational interview (via Vapi) and the post-interview feedback generation (direct API call).

Setup

# 1. Clone & Install
git clone https://github.com/ujjwalv01/IntervU.git && cd IntervU && npm run install:all

# 2. Copy Environment Files (Fill in your keys in both files)
cp client/.env.example client/.env && cp server/.env.example server/.env

# 3. Initialize Database
npm run migrate

# 4. Start Application
npm run dev

The client runs on http://localhost:3000, the server on http://localhost:5001.

Required Accounts

• Neon (neon.tech) — Postgres database, copy the pooled connection string
• Vapi (vapi.ai) — Copy Public Key + Private Key. Groq runs through Vapi's default integration, no extra key needed there
• Groq (console.groq.com) — API key (free tier), used by the server for feedback reports

Optional: Google / GitHub sign-in

Email + password works out of the box. Each social button appears once its keys are in server/.env:

• Google — Google Cloud Console → APIs & Services → Credentials → OAuth client ID (Web application). Redirect URI: http://localhost:5001/api/auth/google/callback
• GitHub — Developer settings → New OAuth App. Callback URL: http://localhost:5001/api/auth/github/callback

In production, set API_URL and CLIENT_URL in server/.env and use your deployed API URL in those callback URLs.

Cost Analysis

Voice is handled by Vapi with Groq as the LLM. Realistic all-in cost is roughly $0.10–0.15/min of conversation (Vapi orchestration + Groq tokens + default STT/TTS), so a 15-minute mock interview costs well under $2. At scale, this is the dominant variable cost of the product and the main lever for margin — cheaper TTS/STT providers or a smaller Groq model would reduce it further at some quality cost.

Key Features

• Adaptive voice interviews - Aria asks follow-up questions based on your actual answers, never from a fixed list
• 4 interview types - Behavioral, Technical, System Design, HR/Culture Fit
• Live transcript - See the conversation as it happens
• Structured feedback - Overall score, category breakdown, strengths, improvements, question-by-question evaluation
• Session history - Dashboard with all past interviews and scores
• Tailored interviews - Paste a job description and add your resume (PDF or text); Aria asks about your real projects and the job's must-haves, and the report rates how well you covered each requirement

Known Limitations / Future Improvements

• Feedback generation is synchronous — adds a few seconds of latency after the call ends. If it fails, the report page offers a one-click retry from the saved transcript.
• Scanned PDF resumes — text is read in the browser, so image-only PDFs need the text pasted instead (no OCR).
• Vapi transcript fallback — if the client-side transcript capture fails, the server reads the transcript from Vapi using the saved call id.
• No admin panel — no way for an evaluator to review interviews or override scores.
• Single user type — no distinction between candidate and recruiter roles.