Quick Start  ·  Features  ·  How It Works  ·  Architecture  ·  API Reference

---

How It Works — Animated

---

Features at a Glance

Voice Interview Engine AI conducts natural voice conversations Deepgram Nova-2 STT (real-time, Indian English) Deepgram Aura TTS (primary) + Edge TTS (free fallback) Domain-aware: Tech, HR, Sales, Ops, CX, PM, Design, Data, Finance Level-calibrated: Intern to Director Time-aware pacing (adapts to interview duration) Custom question banks Resume-based follow-ups

Proctoring & Integrity Face detection (Chrome API + canvas fallback) Eye tracking (gaze direction) Window/app switch detection (3 methods: blur + visibility + focus poll) Phone/device detection (bright object analysis) Mandatory screen sharing + fullscreen enforcement Periodic photo capture (every 60s) Configurable strike system (default 10, server-side count) Copy/paste blocking

Scoring & Analytics Auto-scorecard when interview ends 5 dimensions (Technical, Communication, Problem Solving, Domain, Culture) Evidence-based (exact candidate quotes) Level-calibrated scoring (Intern to Director) STT-aware evaluation (ignores transcription errors) Hire / No Hire recommendation Proctoring report in assessment Candidate comparison (radar chart) DB-backed dedup (no double scoring) Rescore capability

Platform Multi-tenant auth (orgs, roles) Professional dashboard with filters Pagination with ellipsis Question bank management Coding interview mode (Monaco editor) Email notifications Interview recording Resume on reload (state persists) Mobile responsive Docker ready

---

Quick Start

One-command setup

# Clone
git clone https://github.com/vijaygupta18/ai-interview-platform.git
cd ai-interview-platform

# Install
npm install

# Setup database
psql -U postgres -c "CREATE DATABASE ai_interview_platform;"
psql -U postgres -d ai_interview_platform -f migrations/001_schema.sql

# Configure (edit with your API keys)
cp .env.example .env.local

# Start
npm run dev

Open http://localhost:3000 and login with admin@interview.ai / admin123

Docker

docker build -t interview-ai .
docker run -p 3000:3000 --env-file .env.local interview-ai

Environment Variables

Variable	Required	Description
DATABASE_URL	Yes	PostgreSQL connection string
NEXTAUTH_SECRET	Yes	Random secret for JWT signing
NEXTAUTH_URL	Yes	App URL (http://localhost:3000)
AI_BASE_URL	Yes	OpenAI-compatible API base URL
AI_API_KEY	Yes	API key for AI model
AI_MODEL	No	Model name (default: gpt-4o)
DEEPGRAM_API_KEY	Yes	Deepgram API key for STT/TTS
TTS_PROVIDER	No	deepgram (default) or edge (free)
EDGE_TTS_VOICE	No	Voice ID (default: en-IN-NeerjaNeural)
EDGE_TTS_RATE	No	Speed (default: +10%)
SMTP_HOST	No	Email SMTP host
SMTP_PORT	No	Email SMTP port
SMTP_USER	No	Email username
SMTP_PASS	No	Email password

Supported AI Providers

Works with any OpenAI-compatible API:

Provider	AI_BASE_URL	AI_MODEL
OpenAI	https://api.openai.com	gpt-4o, gpt-4o-mini
Anthropic (via proxy)	Your proxy URL	claude-3-5-sonnet
Groq	https://api.groq.com/openai	llama-3.1-70b
Together AI	https://api.together.xyz	meta-llama/Llama-3-70b
Local (Ollama)	http://localhost:11434	llama3
Any OpenAI-compatible	Your endpoint	Your model

---

Pages

Route	Who	Description
/	Interviewer	Dashboard — interviews, filters, pagination, scoring
/new	Interviewer	Create interview — resume, questions, context
/questions	Interviewer	Question bank management
/compare	Interviewer	Side-by-side candidate comparison
/dashboard/[id]	Interviewer	Detail — transcript, scores, photos, proctoring
/review/[id]	Interviewer	Scorecard with score rings
/login	Public	Sign in
/register	Public	Create account + org
/interview/[id]	Candidate	Live interview room (dark theme)
/completed/[id]	Candidate	Thank you page

---

API Reference

Interview Lifecycle

Method	Route	Auth	Description
POST	/api/create-interview	Session	Create interview (FormData)
GET	/api/interview/[id]	Session/Token	Get interview details
POST	/api/interview/[id]/start	Token	Mark started
POST	/api/interview/[id]/end	Token	End + auto-score
GET	/api/interviews	Session	List all (org-scoped)

AI & Speech

Method	Route	Description
POST	/api/ai-speak	Combined AI response + TTS audio
POST	/api/ai-response	AI response text only
POST	/api/tts	Text-to-speech
GET	/api/deepgram-token	Temporary scoped STT token

Scoring & Proctoring

Method	Route	Description
POST	/api/scorecard	Generate/regenerate scorecard
GET	/api/scoring-status/[id]	Check generation status
POST	/api/proctor-event	Log event + photo

Content & Auth

Method	Route	Description
GET/POST	/api/questions	Question bank CRUD
GET/PUT/DELETE	/api/questions/[id]	Single question bank
POST	/api/upload-recording	Upload audio
GET	/api/recording/[id]	Stream recording
POST	/api/auth/register	Create account
GET	/api/health	Health check

---

Database

8 tables — full schema in migrations/001_schema.sql:

organizations ─────── users
       │                 │
       │                 │ (created_by)
       │                 │
       └──── interviews ─┤
              │          │
              │          ├── transcript_entries
              │          ├── proctoring_events (+ photos)
              │          └── interview_rounds
              │
              └── question_banks

webhooks (org-scoped event notifications)

---

TTS Voice Configuration

Provider	Setting	Voice	Cost
Deepgram Aura	TTS_PROVIDER=deepgram	aura-angus-en (Indian male)	$200 free credits
Edge TTS	TTS_PROVIDER=edge	en-IN-NeerjaNeural (Indian female)	Free forever

Indian voices available with Edge TTS:

• en-IN-NeerjaNeural — Professional female
• en-IN-NeerjaExpressiveNeural — Animated female
• en-IN-PrabhatNeural — Professional male
• hi-IN-SwaraNeural — Hindi accent female

---

Security

Protection	Implementation
SQL Injection	Parameterized queries ($1, $2) everywhere
Auth	NextAuth JWT + token validation on all endpoints
Password	bcrypt (cost 12) + min 8 chars server-side
API Keys	Temporary scoped tokens, no main key leak
File Upload	10MB limit + MIME type validation
Path Traversal	UUID regex validation on file paths
XSS	HTML-escaped email templates
Command Injection	execFileSync with array args (no shell)
Rate Limiting	Per-IP limits on critical endpoints
Scoring Dedup	DB-backed atomic lock (survives restart)
Tenant Isolation	Org-scoped queries on all data

---

License

MIT

---