InterviewAI

AI-powered technical interview simulator — practice coding interviews, behavioral sessions, and live collaborative mocks with real-time AI feedback, rubric scoring, and in-depth performance analytics. Fully local, no cloud AI subscriptions required.

---

Table of Contents

1. Features
2. Tech Stack
3. Architecture
4. Getting Started
5. AI Setup
6. Project Structure
7. Key Workflows
8. API Reference
9. Security

---

Features

AI Mock Interviews

• Streaming AI responses that feel like a real interviewer
• Configurable: job title, company, type (behavioral / technical / system design / coding), difficulty (junior / mid / senior), duration
• AI asks follow-up questions based on your answers
• Rubric scoring — every answer graded across Communication, Technical Depth, Problem Solving, and Clarity with a 1–10 score per criterion
• Adaptive questioning — RAG-based question selection prioritises topics you've historically scored low on
• Fallback question pool when AI is unreachable (no silent failures)

Live Code Editor

• Monaco Editor (VS Code engine) with syntax highlighting for 15+ languages
• In-browser code execution via Piston API — Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more
• AI reviews your code solution and explains edge cases
• Dedicated /code-problems page with curated LeetCode-style problems

Collaborative Interview Room

• Share a 6-character room code to join a live session with another person
• Real-time cursor sync, chat, and shared code editor over Socket.IO
• Role-based: one person as interviewer, one as candidate
• AI suggestion panel for interviewers (question hints based on context)

Resume Analyzer

• Paste resume + job description → instant ATS compatibility score
• Matched vs. missing skills breakdown
• Tailored recommendations and predicted interview questions
• One-click: start a mock interview using the generated questions

Analytics Dashboard

• Session history with filtering by type, difficulty, and date
• Score trends over time (bar chart, Y-axis 0–100)
• Breakdown by interview type and difficulty distribution
• Strongest and weakest topic identification

Gamification

• XP system — earn points per session, per completed question, per streak day
• Achievement badges — unlock for milestones (first interview, 7-day streak, perfect rubric score, etc.)
• Daily streak tracking

Additional

• Voice input — Web Speech API for speaking answers (speech-to-text in real time)
• PDF report export — downloadable report per session (pdfkit, server-rendered)
• PWA / Offline — installable app, service worker caches static assets
• Question Bank — browse, filter, and fork community question templates
• Dark / Light mode — system-preference aware

---

Tech Stack

Frontend

Layer	Technology
Framework	React 18 + TypeScript
Build	Vite
UI Components	shadcn/ui + Radix UI primitives
Styling	Tailwind CSS
Code Editor	Monaco Editor
Routing	React Router v6
Charts	Recharts
PWA	vite-plugin-pwa + Workbox
Voice	Web Speech API

Backend

Layer	Technology
Runtime	Node.js (ESM)
Framework	Express.js
AI / LLM	Local LLM engine (configurable model)
Database	SQLite via better-sqlite3 (WAL mode)
Real-time	Socket.IO
Auth	JWT + bcrypt
Code Execution	Piston API (sandboxed)
PDF Export	pdfkit
Security	Helmet, express-rate-limit, CORS

---

Architecture

┌────────────────────────────────────────────────────────────────┐
│                        Browser  (PWA)                          │
│                                                                │
│   ┌───────────────┐   ┌──────────────────┐   ┌─────────────┐  │
│   │  React + UI   │   │  Monaco Editor   │   │ Web Speech  │  │
│   │  (shadcn/ui)  │   │  (code editing)  │   │  API (STT)  │  │
│   └──────┬────────┘   └────────┬─────────┘   └──────┬──────┘  │
└──────────┼─────────────────────┼────────────────────┼─────────┘
           │ REST / SSE stream   │ code run            │ transcript
           ▼                     ▼                     │
┌──────────────────────────────────────────────────────▼─────────┐
│                     Express.js  API  Server                    │
│                                                                │
│   ┌─────────────┐  ┌──────────────┐  ┌──────────────────────┐  │
│   │  /api/ai    │  │  /api/code   │  │  /api/resume         │  │
│   │  streaming  │  │  execution   │  │  /api/interviews      │  │
│   │  responses  │  │  (Piston)    │  │  /api/reports        │  │
│   └──────┬──────┘  └──────┬───────┘  └──────────┬───────────┘  │
│          │                │                      │              │
│   ┌──────▼──────┐  ┌──────▼───────┐  ┌──────────▼───────────┐  │
│   │  Local LLM  │  │  Piston API  │  │  SQLite (WAL mode)   │  │
│   │   Engine    │  │  (external)  │  │  sessions · users    │  │
│   │ (local      │  │              │  │  questions · reports │  │
│   │  inference) │  └──────────────┘  └──────────────────────┘  │
│   └─────────────┘                                              │
│                                                                │
│   ┌────────────────────────────────────────────────────────┐   │
│   │          Socket.IO  —  Collaborative Interviews        │   │
│   │   room creation · cursor sync · shared code editor    │   │
│   └────────────────────────────────────────────────────────┘   │
└────────────────────────────────────────────────────────────────┘

Data Flow — Interview Session

User opens setup page
        │
        ▼
POST /api/ai/generate-question
        │
        ▼
AI engine streams question tokens → SSE → browser renders word-by-word
        │
        ▼
User types / speaks answer
        │
        ▼
POST /api/ai/interview  (answer + context)
        │
        ├──▶ AI engine: generate follow-up + rubric JSON
        │
        └──▶ SQLite: save message + rubric scores
                │
                ▼
        Rubric panel updates live (Communication / Technical / etc.)
                │
                ▼
        Session end → POST /api/reports → pdfkit → PDF download

Database Schema

users
  id · email · passwordHash · xp · streak · lastActive

sessions
  id · userId · config(JSON) · messages(JSON) · scores(JSON)
  totalScore · duration · completedAt

question_bank
  id · title · content · type · difficulty · tags
  authorId · forkCount · usageCount

question_embeddings
  questionId · embedding(BLOB) · model

---

Getting Started

Prerequisites

• Node.js 18+
• A local LLM engine installed and running (see AI Setup)

Install & Run

# 1. Clone
git clone https://github.com/SriramAtmakuri/InterviewAI
cd InterviewAI

# 2. Install frontend
npm install

# 3. Install backend
cd backend && npm install && cd ..

# 4. Pull your preferred AI model (see AI Setup below)

# 5. Start backend
cd backend && npm run dev &

# 6. Start frontend
npm run dev

---

AI Setup

InterviewAI runs AI inference locally — no API keys, no usage fees, no data sent to external services. Configure your preferred local LLM engine in backend/.env:

AI_BASE_URL=http://your-ai-engine:port
AI_MODEL=your-preferred-model

The backend health endpoint (/health) reports AI connection status and lists available models on startup.

---

Project Structure

InterviewAI/
├── src/                          # Frontend (React)
│   ├── pages/
│   │   ├── Landing.tsx           # Marketing / hero page
│   │   ├── Auth.tsx              # Login / register
│   │   ├── Dashboard.tsx         # Home after login
│   │   ├── InterviewSetup.tsx    # Configure interview
│   │   ├── InterviewSession.tsx  # Active interview + rubric
│   │   ├── CodeProblems.tsx      # LeetCode-style problems
│   │   ├── History.tsx           # Past sessions
│   │   ├── Analytics.tsx         # Charts + metrics
│   │   ├── ResumeAnalyzer.tsx    # ATS + skill gap
│   │   ├── CollabInterview.tsx   # Live collaborative room
│   │   └── QuestionBank.tsx      # Community templates
│   ├── components/
│   │   ├── CodeEditor.tsx        # Monaco wrapper
│   │   ├── GamificationPanel.tsx # XP + badges
│   │   ├── InterviewHints.tsx    # AI hint system
│   │   └── AIStatusBadge.tsx     # AI engine health indicator
│   └── config/
│       └── ollama.ts             # AI backend URL config
│
├── backend/
│   ├── src/
│   │   ├── index.js              # Express server + Socket.IO init
│   │   ├── config/
│   │   │   ├── database.js       # SQLite setup + migrations
│   │   │   └── ollama.js         # AI client + health check
│   │   ├── routes/               # Express route definitions
│   │   ├── controllers/          # Request handlers
│   │   ├── services/
│   │   │   └── ollamaService.js  # AI prompting + RAG logic
│   │   ├── middleware/
│   │   │   ├── auth.js           # JWT verify
│   │   │   └── rateLimiter.js    # Per-route rate limits
│   │   └── wsHandlers.js         # Socket.IO collab logic
│   └── .env                      # Environment config
│
└── demo-recording/
    └── interviewai.gif           # Feature walkthrough

---

Key Workflows

Starting a Mock Interview

1. Dashboard → Start Mock Interview
2. Set job title, type (behavioral / technical / system design), difficulty, duration
3. Optionally add custom questions or import from Resume Analyzer
4. Session starts → AI streams first question
5. Answer via text or voice → AI follows up → rubric updates per response
6. End session → view detailed scoring → download PDF report

Collaborative Interview

1. One person creates a room → shares the 6-character code
2. Interviewer and candidate join → both see the shared code editor
3. Interviewer receives AI question suggestions in a side panel
4. Session log saved to both participants' history

Resume → Interview Pipeline

1. Resume Analyzer → paste resume + job description
2. View ATS score, matched/missing skills, recommendations
3. Click Practice These Now → launches interview pre-loaded with predicted questions

---

API Reference

Auth

Method	Endpoint	Description
POST	/api/auth/register	Create account
POST	/api/auth/login	Get JWT token
POST	/api/auth/refresh	Rotate refresh token
POST	/api/auth/logout	Revoke refresh token
GET	/api/auth/me	Current user profile

AI / Interviews

Method	Endpoint	Description
POST	/api/ai/interview	Send answer, stream AI response
POST	/api/ai/generate-question	Get opening question for config
POST	/api/ai/analyze-resume	ATS score + skill gap
POST	/api/ai/validate-code	AI code review

Sessions & Reports

Method	Endpoint	Description
POST	/api/interviews	Save completed session
GET	/api/interviews	List user sessions
GET	/api/interviews/:id	Session detail + rubric
GET	/api/reports/:id	Download PDF report

Code Execution

Method	Endpoint	Description
POST	/api/code/execute	Run code (Piston sandbox)
GET	/api/code/languages	List supported languages

Question Bank

Method	Endpoint	Description
GET	/api/templates	Browse community questions
POST	/api/templates	Publish a question template
POST	/api/templates/:id/fork	Fork to your bank

---

Security

• JWT authentication — short-lived access tokens + rotating refresh tokens, signed with a 256-bit secret
• bcrypt password hashing — cost factor 10
• Rate limiting — per-route limits (30 AI req/min, 10 code exec/min, 100 general/min)
• Helmet.js — secure HTTP headers (CSP, HSTS, X-Frame-Options, etc.)
• CORS — restricted to configured frontend origin
• Input validation — all API boundaries validated before processing
• Code sandboxing — code execution runs in Piston's isolated containers, never on the host
• Local AI — all inference runs on-device; no prompts or responses leave your machine