JobSense AI — Autonomous Job Hunt Agent

A personal automation agent that scrapes job portals every morning, scores each listing against my resume using Gemini AI, and delivers a ranked WhatsApp digest and email report — automatically, every day.

Built because I was tired of manually checking 4 job boards every morning.

---

How It Works

QStash (daily schedule)
    ↓
Express Webhook
    ↓
Scrapers (Adzuna · Remotive · Arbeitnow)
    ↓
Redis Deduplication (Upstash · 7-day TTL)
    ↓
Gemini AI Pipeline
    ├── scoreJob              → match score, skill gaps, salary fit
    ├── assessResumeForJob    → hiring manager feedback, project to highlight
    └── rewriteBulletsForJob  → keyword-optimized resume bullets
    ↓
PostgreSQL (Neon · via Prisma)
    ↓
Notifications
    ├── WhatsApp digest (Twilio)
    └── Email report (Nodemailer)

---

Tech Stack

Technology	Purpose
Node.js 22	Runtime
TypeScript	Language
Express.js	Web framework
PostgreSQL (Neon)	Primary database
Prisma ORM	Database access
Redis (Upstash)	Job deduplication cache
Gemini AI	Scoring, assessment, bullet rewriting
QStash	Daily scheduling and webhook delivery
Twilio WhatsApp	Morning digest notification
Nodemailer	Email report
Arcjet	Rate limiting and bot protection
Docker Compose	Local development environment
Koyeb	Cloud deployment

---

AI Pipeline Detail

Each new job goes through up to 3 Gemini AI steps:

1. scoreJob — runs on all new jobs Scores the job 0–100 against the candidate resume. Returns:

• Match level: STRONG / GOOD / WEAK / SKIP
• Matching and missing skills
• Experience gap analysis
• Salary fit assessment
• Why to apply

2. assessResumeForJob — STRONG matches only Hiring manager perspective on the candidate for this specific role. Returns:

• Overall verdict
• Strengths and weaknesses
• Which project to highlight and why
• Keywords to add to resume
• Personalized cover letter hook
• Projected score if gaps are fixed

3. rewriteBulletsForJob — STRONG matches only Rewrites resume project bullets to be metric-driven and keyword-optimized for this specific job. Only uses skills already on the resume — no fabrication.

---

Project Structure

src/
├── ai/
│   ├── client.ts         # Gemini client with retry logic
│   ├── resume.ts         # Candidate resume as structured constant
│   ├── scorer.ts         # Job scoring prompt
│   ├── assessor.ts       # Resume assessment prompt
│   └── improver.ts       # Bullet rewriter prompt
├── cache/
│   └── redis.ts          # Upstash deduplication with 7-day TTL
├── db/
│   ├── jobs.ts           # Save scored jobs to PostgreSQL
│   └── runs.ts           # Daily run tracking
├── middleware/
│   ├── arcjet.ts         # Rate limiting and bot protection
│   └── qstash.ts         # QStash signature verification
├── notifications/
│   ├── whatsapp.ts       # Twilio WhatsApp digest
│   └── email.ts          # Nodemailer email report
├── routes/
│   └── trigger.ts        # Main pipeline route
├── scrapers/
│   ├── adzuna.ts         # Adzuna Jobs API
│   ├── remotive.ts       # Remotive API
│   ├── arbeitnow.ts      # Arbeitnow API
│   └── index.ts          # Run all scrapers
└── app.ts                # Express app entry point

---

Quick Start

Prerequisites

• Node.js 22+
• Neon PostgreSQL database
• Upstash Redis database
• Gemini API key (free tier)
• QStash account (free tier)
• Twilio account (for WhatsApp)
• Gmail account (for email reports)

Setup

# Clone the repo
git clone https://github.com/yourusername/jobsense-ai
cd jobsense-ai

# Install dependencies
npm install

# Copy environment variables
cp .env.example .env
# Fill in your values

# Run database migrations
npx prisma migrate deploy

# Start development server
npm run dev

Trigger the pipeline manually

curl -X POST http://localhost:3000/webhook/trigger

---

Environment Variables

PORT=3000
NODE_ENV=development

# Database
DATABASE_URL=postgresql://...

# Redis
REDIS_URL=rediss://...

# AI
GEMINI_API_KEY=your_key

# QStash
QSTASH_TOKEN=your_token
QSTASH_CURRENT_SIGNING_KEY=your_key
QSTASH_NEXT_SIGNING_KEY=your_key

# Twilio WhatsApp
TWILIO_ACCOUNT_SID=ACxxx
TWILIO_AUTH_TOKEN=your_token
TWILIO_WHATSAPP_FROM=whatsapp:+14155238886
YOUR_WHATSAPP_NUMBER=whatsapp:+91XXXXXXXXXX

# Email
GMAIL_USER=your@gmail.com
GMAIL_APP_PASSWORD=your_app_password
EMAIL_TO=your@gmail.com

# Security
ARCJET_KEY=ajkey_xxx

---

API Endpoints

Method	Endpoint	Description
POST	/webhook/trigger	QStash daily pipeline trigger
GET	/health	Service health check
GET	/jobs	View all scored jobs

---

Deployment

Deployed on Koyeb (free tier) with:

• Neon for PostgreSQL
• Upstash for Redis
• QStash for daily scheduling

# Build
npm run build

# Start
node dist/app.js

Update your QStash webhook URL to your Koyeb deployment URL after deploying.

---

Design Decisions

Why APIs instead of scraping? Instahyre, Wellfound, and Cutshort all use Cloudflare bot protection — Axios + Cheerio returns undefined before parsing anything. Switching to open APIs made scrapers reliable and maintainable.

Why hardcode the resume instead of uploading a PDF? This is a personal automation tool — the resume doesn't change daily. A structured TypeScript constant gives Gemini clean, parseable context on every API call without needing file upload infrastructure.

Why score only 10 jobs per run? Gemini free tier has a daily quota. Scoring all 140 jobs × 3 AI calls = 420 API calls per run. Limiting to top 10 new jobs keeps usage within free tier permanently.

---

Author

Konduru Durga Prasad Full Stack Developer · Backend Engineer · Bangalore, India

GitHub · LinkedIn · kondurudurgaprasad.2@gmail.com