README.md

not your average resume tool. this one actually works.

---

okay real talk.

you applied to 47 jobs last month.

got 2 callbacks.

spent 6 hours "tailoring" resumes that were basically the same doc
with a different company name pasted at the top.

the ATS ate them all. silently. without hesitation.

this backend was built to fix that.

AI Resume Tailor is a production-grade SaaS backend that generates ATS-optimized resumes using Gemini AI, scores them, exports them as PDFs, and runs on a real coin-based payment system powered by Razorpay.

it's not a side project. it's a startup backend dressed as one.

---

the stack (no bloat, only facts)

Layer	Tech	Why
Runtime	Node.js + Express.js	fast, scalable, industry standard
Database	MongoDB + Mongoose	flexible schema for resume data
AI Engine	Google Gemini API	best-in-class generation quality
Payments	Razorpay	UPI + cards + wallets, built for India
PDF Export	LaTeX + latexmk	pixel-perfect, recruiter-ready output
Auth	JWT	stateless, secure, no session headaches

---

what it actually does

AI Resume Generation — drop your CV, paste the JD, get a tailored resume in seconds

feeds your resume, the job description, and your LaTeX template into Gemini with a smart prompt that forces it to:

• inject ATS keywords from the JD directly
• swap weak verbs for strong ones ("worked on" → "engineered", "helped" → "led")
• add measurable metrics where they're missing
• keep your LaTeX structure completely intact

no hallucinated experience. no generic filler. targeted output only.

Prompt Optimization Engine — the brain behind the brain

before anything reaches Gemini, the prompt engine does its own analysis:

detects your target role  →  Frontend / Backend / DevOps / etc.
extracts JD keywords      →  maps them to resume sections
identifies gaps           →  flags what's missing vs. what's needed
builds the prompt         →  few-shot optimized for high-quality output
sends to Gemini           →  now the AI actually knows what to do

this is why the output doesn't sound like generic ChatGPT resume advice.

ATS Score API — know your number before you apply

{
  "score": 84,
  "keyword_match": "79%",
  "matched": ["React", "Node.js", "REST API", "MongoDB", "AWS"],
  "missing": ["TypeScript", "Docker", "CI/CD", "Jest", "Redis"],
  "suggestions": [
    "Add TypeScript to skills — appears 6x in JD",
    "Mention Docker in at least one project bullet",
    "CI/CD experience is required, not preferred — address it explicitly"
  ]
}

no more guessing. hard numbers. specific fixes.

PDF Generation — LaTeX → clean PDF → download → done

converts your optimized LaTeX resume to a properly formatted PDF using latexmk. temp files are handled securely and cleaned up post-download. no leftover junk, no exposed paths.

Coin System + Razorpay Payments — real monetization, real logic

users buy coins. coins unlock generations. coins only deduct after a successful AI response. no charge for failures.

user buys 100 coins via Razorpay
         ↓
payment verified via signature check (no fake webhooks)
         ↓
coins credited to account atomically
         ↓
generate resume = spend coins
         ↓
only deducted if Gemini returns valid output

double-crediting protection built in. production-safe.

Auth + Security — not an afterthought

• JWT authentication on all protected routes
• rate limiting to block abuse
• input validation before anything hits the DB or AI
• secure temp file handling for LaTeX compilation

---

the full request flow

POST /api/generate/generate
         │
         ▼
  authMiddleware.js          ← is this token valid?
         │
         ▼
  rateLimiter.js             ← is this user spamming?
         │
         ▼
  generateController.js      ← does the user have coins?
         │
         ▼
  ats.js                     ← run ATS pre-analysis
         │
         ▼
  promptEngine.js            ← build optimized Gemini prompt
         │
         ▼
  gemini.js                  ← call Gemini API
         │
         ▼
  postProcess()              ← clean + validate output
         │
         ▼
  Resume.save()              ← store to MongoDB
  User.deductCoins()         ← only now, coins go
         │
         ▼
  return optimized resume    ← 200. you're welcome.

---

project structure

server/
│
├── controllers/
│   ├── authController.js       # register, login, token handling
│   ├── generateController.js   # resume generation + ATS scoring
│   ├── paymentController.js    # Razorpay order creation + verification
│   └── userController.js       # profile, coin balance, history
│
├── models/
│   ├── User.js                 # user schema + coin balance
│   ├── Resume.js               # resume history per user
│   └── Payment.js              # payment records + idempotency
│
├── routes/
│   ├── authRoutes.js
│   ├── generateRoutes.js
│   └── paymentRoutes.js
│
├── middleware/
│   ├── authMiddleware.js       # JWT verification
│   └── rateLimiter.js          # express-rate-limit config
│
├── utils/
│   ├── gemini.js               # Gemini API wrapper
│   ├── ats.js                  # ATS scoring engine
│   ├── promptEngine.js         # smart prompt builder
│   ├── pdfGenerator.js         # LaTeX → PDF pipeline
│   └── razorpay.js             # payment utils + signature verify
│
├── .env
├── app.js
└── server.js

---

API reference

Auth

POST   /api/auth/register          →  create account
POST   /api/auth/login             →  get JWT token

Resume

POST   /api/generate/generate      →  generate tailored resume (costs coins)
POST   /api/generate/ats-score     →  score resume against JD (free)
POST   /api/generate/download-pdf  →  compile + download PDF
GET    /api/generate/history       →  past generated resumes

Payments

POST   /api/payment/create-order   →  create Razorpay order
POST   /api/payment/verify         →  verify signature + credit coins

---

setup

1. clone and install

git clone https://github.com/anubhavxdev/ai-resume-tailor-backend.git
cd ai-resume-tailor-backend
npm install

2. environment variables

# create .env in root
PORT=5000
MONGO_URI=your_mongodb_connection_string
JWT_SECRET=something_long_and_random
GEMINI_API_KEY=your_gemini_api_key
RAZORPAY_KEY_ID=your_razorpay_key
RAZORPAY_KEY_SECRET=your_razorpay_secret

3. install LaTeX (required for PDF generation)

# macOS
brew install --cask mactex

# Ubuntu / Debian
sudo apt-get install texlive-full

# Windows
# install MiKTeX from https://miktex.org

# verify
latexmk -v

4. run

npm run dev

---

production checklist

before you ship this, make sure you've handled:

• Razorpay webhooks for payment reliability (not just verify endpoint)
• MongoDB transactions for atomic coin operations
• Winston or Morgan for structured logging
• Nginx reverse proxy + PM2 process manager
• ATS result caching (Redis) — Gemini calls aren't free
• Environment secrets via AWS Secrets Manager or similar
• Health check endpoint for uptime monitoring
• Rate limiting tuned per route, not global

---

roadmap

• Gemini-powered resume generation
• ATS scoring engine
• Prompt optimization layer
• LaTeX → PDF pipeline
• Razorpay payment + coin system
• JWT auth + rate limiting
• Resume improvement loop (generate → score → improve → repeat)
• Multi-step AI pipeline with intermediate validation
• AI resume feedback assistant (conversational)
• Analytics dashboard for usage + revenue
• React + Tailwind frontend
• Chrome extension for one-click JD import

---

what makes this different

most "AI resume" projects are a Gemini API call wrapped in a POST /generate route.

this one has a prompt engine, an ATS layer, a payment system, PDF compilation, and a coin economy — all wired together with proper middleware, error handling, and production-aware logic.

it's not built to look good in a portfolio.

it's built to run in prod.

---

Anubhav Jaiswal B.Tech CS · Lovely Professional University Founder @ FounDev Studio · Tech Lead @ AWS Cloud Community · 2x Hackathon Winner

---

if this hit different —