InterviewPilot AI

AI-powered voice interviews that actually prepare you for the real thing.

Live Demo · GitHub

---

Overview

InterviewPilot AI is a production-grade voice interview platform. It simulates realistic technical and behavioral interviews using ElevenLabs ConvAI for natural speech, a deterministic conversation engine for adaptive follow-ups, and GPT-4 for detailed feedback — backed by PostgreSQL, Redis, and a clean layered architecture.

Unlike static question banks, InterviewPilot:

• Listens to your actual spoken answers
• Adapts with contextual follow-up questions
• Remembers your resume, job description, and previous answers
• Challenges weak reasoning and probes trade-off decisions
• Evaluates with recruiter-quality feedback and scores

---

Screenshots

Dashboard & Setup

Interview Experience

Platform

---

Key Features

Voice AI

• ElevenLabs ConvAI (primary) with OpenAI Realtime fallback
• Real-time Web Audio API waveform visualization
• Voice Activity Detection with silence/speaking/loud states
• Barge-in — interrupt the AI mid-response
• Streaming transcript with speaker identification

Conversation Engine

• 12-state deterministic machine — no LLM owns the interview flow
• 8 interview types — Behavioral, Technical, System Design, Frontend, Backend, Full-Stack, DevOps, Mixed
• Adaptive difficulty — scales from foundational to principal based on response quality
• 30+ follow-up styles — never repeats phrasing
• Context-aware — uses resume, job description, and conversation history

Feedback

• Overall score with Communication, Confidence, and Technical Reasoning breakdowns
• Hiring recommendation — Strong Hire / Hire / Lean Hire / No Hire
• Strengths, weaknesses, and actionable improvement plan
• GPT-4 powered analysis with transcript citations

Platform

• JWT auth (15min access + 7d refresh) in httpOnly cookies
• bcrypt (cost 12) + CSRF + rate limiting
• PostgreSQL via Prisma ORM with Redis cache-aside
• Startup health checks for DB, Redis, OpenAI, and every ElevenLabs agent
• Dark mode, responsive, accessible

---

Architecture

Browser (Web Audio API + WebSocket)
   │
   ├── getUserMedia → Shared AudioContext (16kHz)
   │   ├── AnalyserNode → VAD → Waveform
   │   └── ScriptProcessor → PCM16 → WebSocket
   │
   ▼
Next.js 16 App Router
   │
   ├── proxy.ts — Security headers, auth redirects
   ├── Route Handlers (/api/v1/*) — Zod validation
   ├── Service Layer — Auth + Interview
   ├── Repository Layer — Prisma with cache-aside
   │
   ├── AI Provider Abstraction
   │   ├── ElevenLabs ConvAI (primary)
   │   ├── OpenAI (fallback + feedback)
   │   └── Agent Pool — weighted random + circuit breaker
   │
   ├── Conversation Engine
   │   ├── State machine (12 states)
   │   ├── Context engine (profile, resume, JD, history)
   │   ├── Prompt engine (modular, 8 templates)
   │   └── Token manager (100K budget)
   │
   └── Monitoring — Pino + Startup health checks

---

Tech Stack

Category	Technology
Framework	Next.js 16, React 19, TypeScript
Styling	TailwindCSS, shadcn/ui, Framer Motion
Backend	Next.js Route Handlers
Database	PostgreSQL (Supabase)
ORM	Prisma 6
Cache	Redis (ioredis) + Memory fallback
Voice AI	ElevenLabs ConvAI, OpenAI Realtime
Text AI	GPT-4 / GPT-4.1
Auth	JWT (jose), bcrypt, SHA-256
Validation	Zod v4
Logging	Pino v10
Deployment	Vercel

---

Interview Flow

Candidate → Setup → Resume Upload → Voice Session
    │
    ├── AI greets by name, references experience
    ├── Adaptive questions based on type, level, JD
    ├── Follow-ups probe depth, challenge reasoning
    ├── Barge-in for natural turn-taking
    ├── Live transcript streaming
    │
    ▼
Interview Ends → GPT-4 Feedback → Report → Dashboard

---

Getting Started

git clone https://github.com/niranjansah87/interviewpilot-ai.git
cd interviewpilot-ai
pnpm install
cp .env.example .env.local
pnpm prisma migrate dev
pnpm dev

---

Environment Variables

Variable	Required	Description
DATABASE_URL	Yes	PostgreSQL connection (Supabase)
JWT_SECRET	Yes	Access token signing (min 32 chars)
JWT_REFRESH_SECRET	Yes	Refresh token signing (min 32 chars)
OPENAI_API_KEY	Yes	OpenAI API key
NEXT_PUBLIC_APP_URL	Yes	Application URL
ELEVENLABS_API_KEY	Voice	Primary ElevenLabs key
ELEVENLABS_AGENT_ID	Voice	Primary ConvAI agent
ELEVENLABS_BACKUP_API_KEY	Optional	Backup agent key
ELEVENLABS_BACKUP_AGENT_ID	Optional	Backup agent ID
ELEVENLABS_EXTRA_API_KEY	Optional	Extra agent key
ELEVENLABS_EXTRA_AGENT_ID	Optional	Extra agent ID
REDIS_URL	Optional	Redis connection
CACHE_PROVIDER	Optional	redis or memory (default)

---

Project Structure

src/
├── app/                     # Next.js App Router
│   ├── (auth)/              # Login, register
│   ├── (dashboard)/         # Protected routes
│   └── api/v1/              # REST API (auth, users, interviews, voice)
├── components/
│   ├── ui/                  # shadcn/ui components
│   └── features/            # Header, Sidebar, VoiceInterface
├── hooks/                   # useInterviewSession
├── lib/
│   ├── ai/                  # Provider abstraction, agent pool, adapters
│   ├── auth/                # JWT, bcrypt, cookies, CSRF
│   ├── conversation/        # Engine, context, prompts, token manager
│   ├── api/                 # Route helpers, rate limiting
│   └── audio/               # Shared AudioContext runtime
├── services/                # Auth + interview business logic
├── repositories/            # Prisma data access with cache-aside
├── cache/                   # CacheProvider, Redis, Memory
├── config/                  # Zod-validated environment
├── proxy.ts                 # Security headers + auth redirects
└── styles/                  # Design tokens (light + dark)

---

Design Decisions

Decision	Rationale
Next.js App Router	Server Components, Route Handlers, single deployment
ElevenLabs ConvAI	Production voice quality, WebSocket streaming, VAD built-in
Provider abstraction	Swap voice providers without touching application code
Agent pool with circuit breaker	Load-balance across 3 agents, auto-skip failed ones
Shared AudioContext	Single source for mic, VAD, and visualization — zero latency
Prisma + PostgreSQL	Type-safe queries, migrations, Supabase managed hosting
Redis cache-aside	Automatic memory fallback, no single point of failure
Deterministic conversation engine	LLM generates language, engine owns the interview flow

---

API Endpoints

Method	Route	Description
POST	/api/v1/auth/register	Create account
POST	/api/v1/auth/login	Sign in
POST	/api/v1/auth/logout	Sign out
POST	/api/v1/auth/refresh	Refresh tokens
GET	/api/v1/users/me	Get profile
PATCH	/api/v1/users/me/name	Update name
POST	/api/v1/users/me/password	Change password
POST	/api/v1/users/me/resume	Upload resume
GET	/api/v1/interviews	List interviews
POST	/api/v1/interviews	Create interview
GET	/api/v1/interviews/[id]	Get interview
PATCH	/api/v1/interviews/[id]	Update status
DELETE	/api/v1/interviews/[id]	Delete interview
POST	/api/v1/interviews/[id]/report	Generate feedback
POST	/api/v1/interviews/[id]/transcript	Save transcript
POST	/api/v1/voice/connect	Get signed ElevenLabs URL
GET	/api/v1/health	Health check

---

Security

• JWT access tokens (15 min) + refresh tokens (7 days) in httpOnly, SameSite cookies
• bcrypt password hashing at cost 12
• CSRF double-submit cookie with AES-256-GCM signatures
• Rate limiting on auth endpoints (10 req / 15 min per IP)
• Zod v4 validation on every API input
• Security headers — CSP, X-Frame-Options, X-Content-Type-Options, Permissions-Policy
• Prisma parameterized queries prevent SQL injection
• API keys never reach the browser — signed URLs generated server-side

---

Diagrams

System architecture, voice pipeline, and interview lifecycle — rendered as Mermaid diagrams on GitHub.

Diagram	Description
System Architecture	Full system overview — browser, Next.js, AI providers, database
Voice Pipeline	Microphone → Web Audio API → VAD → ElevenLabs → Playback
Interview Lifecycle	State machine: idle → connecting → listening → speaking → completed
End-to-End Flow	Complete user journey from registration to feedback report
Database Schema	PostgreSQL tables, relationships, indexes, enums
Auth Flow	JWT login, refresh, CSRF, cookie flow
Provider Architecture	AI provider abstraction, agent pool, circuit breaker
Authentication	Login sequence, token exchange, cookie flow
Deployment	Vercel + Supabase + Redis deployment topology
Interview Engine	Conversation engine state machine and decision flow
Voice Flow	WebSocket audio streaming and transcript pipeline

---

Documentation

• Architecture
• Tech Stack
• AI Engine
• Conversation Engine
• Database
• API
• Security
• Caching
• ADR Documents

---

License

MIT © Niranjan Sah