Prepify 🎯

An AI-powered interview preparation SaaS that simulates structured interviews, evaluates candidate responses with LLMs, and generates comprehensive diagnostic feedback reports.

---

📌 Table of Contents

• Overview & Problem Statement
• System Architecture
• Key Features
• Tech Stack
• Engineering Highlights & Design Decisions
• API Overview
• Getting Started
• Environment Variables
• Author

---

💡 Overview & Problem Statement

• The Problem: Job seekers and developers lack low-stress, interactive environments to practice technical and behavioral interviews. Traditional mock interviews are difficult to schedule, lack standardized scoring, and rarely provide immediate, actionable feedback.
• The Solution: Prepify automates interview evaluation by combining full-stack MERN architecture with large language models. Users complete guided interview sessions and receive structured reports identifying conceptual gaps, communication clarity, and targeted areas for improvement.

---

🏗️ System Architecture

┌────────────────────────────────────────────────────────┐
│                   React + Vite Client                  │
│       (Tailwind CSS, Axios Interceptors)               │
└───────────────────────────┬────────────────────────────┘
                            │
                            │ HTTPS / RESTful API
                            ▼
┌────────────────────────────────────────────────────────┐
│                 Node.js / Express Server               │
│  ┌───────────────────┐        ┌─────────────────────┐  │
│  │  JWT Auth & CORS  │        │ Custom Rate Limiter │  │
│  └─────────┬─────────┘        └──────────┬──────────┘  │
│            │                             │             │
│            ▼                             ▼             │
│  ┌───────────────────┐        ┌─────────────────────┐  │
│  │ Interview Engine  │───────>│  LLM Report Service │  │
│  └─────────┬─────────┘        └──────────┬──────────┘  │
└────────────┼─────────────────────────────┼─────────────┘
             │                             │
             ▼                             ▼
   ┌──────────────────┐          ┌───────────────────┐
   │ MongoDB Database │          │ LLM / AI Provider │
   │ (Mongoose Models)│          │   (OpenAI API)    │
   └──────────────────┘          └───────────────────┘

✨ Key Features

• AI-Driven Performance Reports: Transforms raw interview transcripts and user responses into categorized feedback highlighting strengths, technical depth, and specific blind spots using structured LLM prompts.

• Session & Interview Lifecycle Management: Create, pause, retrieve, and review historical interview sessions backed by persistent MongoDB schemas.

• Stateless Authentication & Resource Isolation: Implements secure JWT/cookie authentication to protect user records, ensuring session reports are private and strictly bound to owner accounts.

• Defensive Rate Limiting: Enforces strict throughput throttling around expensive AI evaluation endpoints to prevent abuse, race conditions, and uncontrolled API cost overruns.

• Responsive, Accessible Interface: Designed with Vite, React, Tailwind CSS, and Shadcn UI components with dynamic feedback using Lucide icons and React Toastify.

🛠️ Tech Stack

Layer	Technologies
Frontend	React 18, Vite, Tailwind CSS, Shadcn UI, Lucide React, Axios, React Toastify
Backend	Node.js, Express.js, REST Architecture
Database	MongoDB, Mongoose ODM
AI Integration	Large Language Model (LLM) Inference APIs (OpenAI / External AI Gateway)
Security & Utilities	JSON Web Tokens (JWT), HTTP-Only Cookies, Express Rate Limit, CORS
Deployment	Vercel (Frontend), Render (Backend), MongoDB Atlas

🧠 Engineering Highlights & Design Decisions

• Cost & Abuse Protection (Rate Limiting):

  • Challenge: Unrestricted access to LLM inference endpoints can lead to denial-of-wallet (DoW) attacks, API abuse, and unnecessary computation costs.
  • Solution: Implemented middleware-level rate limiting on /api/reports/generate to restrict report-generation frequency per user/IP while keeping previously generated reports accessible without triggering new LLM requests.

• Structured LLM Output Validation:

  • Challenge: Non-deterministic LLM responses can produce malformed or inconsistent JSON, potentially breaking frontend rendering and corrupting persisted data.
  • Solution: Added defensive schema validation for AI-generated report payloads before persisting them to MongoDB, ensuring only structurally valid reports reach the database and client.

• Decoupled Client–Server State Management:

  • Challenge: Multi-step interview workflows require reliable handling of asynchronous API operations without blocking or disrupting the user interface.
  • Solution: Separated API communication into modular Axios service layers and managed UI state locally at the component level, enabling predictable state transitions and responsive feedback during background report generation.

  🔌 API Overview

Method	Endpoint	Description	Auth Required
POST	/api/auth/register	Register a new user	No
POST	/api/auth/login	Authenticate user & issue token	No
GET	/api/interviews	List all historical sessions for the active user	Yes
POST	/api/interviews	Initialize a new interview session	Yes
POST	/api/reports/generate	Generate AI report for a completed session (Rate Limited)	Yes
GET	/api/reports/:id	Fetch detailed report by ID	Yes

---

🚀 Getting Started

Prerequisites

Make sure the following are installed:

• Node.js: >= 18.x
• npm: >= 9.x
• MongoDB: Local MongoDB instance or MongoDB Atlas

1. Clone the Repository

git clone https://github.com/abhay-0907/Prepify
cd prepify

Backend Setup:

cd backend
npm install
cp .env.example .env
# Add your environment variables in .env

npm run dev Backend runs on http://localhost:3000 by default.

2. Frontend Setup:

cd ../frontend
npm install
cp .env.example .env
# Set VITE_API_URL in .env
npm run dev
Frontend runs on http://localhost:5173.

⚙️ Environment Variables

Backend (/backend/.env)

PORT=3000
MONGODB_URI=mongodb+srv://<username>:<password>@cluster.mongodb.net/prepify?retryWrites=true&w=majority

JWT_SECRET=your_jwt_secret_key

OPENAI_API_KEY=your_openai_api_key

FRONTEND_URL=http://localhost:5173

Frontend (/frontend/.env)