🤖 HireAgent

AI-Powered Job Discovery, Resume Tailoring & Auto-Application Pipeline

Built by Gunakarthik Naidu Lanka MS Computer Science @ Arizona State University • GPA 4.0

From job boards → tailored resume → submitted application. Fully automated.

---

🚀 What Is HireAgent?

HireAgent is a production-grade agentic pipeline that automates the entire job hunt for entry-level software engineers.

It scrapes 8+ job boards, scores every role with an LLM, generates a unique tailored one-page resume (LaTeX → PDF) per job, and fills out ATS application forms using a Vision-Verified Browser Agent — all from a single terminal command.

---

⚡ How It Works — The Pipeline

Stage	What Happens	Model Used
🔍 Discover	Scrapes 8 job boards for entry-level SWE roles	—
📋 Enrich	Visits each job page, pulls full description + apply URL	Ollama gemma3:4b
🧠 Score	Rates each job 1–10 (relevance, tech stack)	NVIDIA DeepSeek-V3
✏️ Tailor	Rewrites resume bullets to mirror the JD language	NVIDIA DeepSeek-R1
📄 Cover	Generates a targeted cover letter	NVIDIA DeepSeek-R1
🖨️ PDF	Compiles tailored resume → one-page PDF via LaTeX	—
🖱️ Apply	Vision-verified fill + Enterprise CAPTCHA Solving	NVIDIA Llama-3.2-Vision

---

🦾 Autonomous Application Engine (Apply)

The application engine has been upgraded for maximum resilience:

1. Vision-Verified Filling (SoM)

Using a Set-of-Marks (SoM) approach, HireAgent takes a screenshot of the form, draws bounding boxes around every interactive element, and uses Llama-3.2-11B-Vision to perfectly map your profile data to the visual layout.

2. Enterprise CAPTCHA Solving

Integrated with CapSolver API to automatically solve:

• 🧩 hCaptcha: Specialist detection for Lever/Veeva formats.
• 🌀 Cloudflare Turnstile: Handles script & iframe injected challenges.
• 🤖 reCAPTCHA v2 / Enterprise: Sitekey-based solving and injection.
• 🛡️ Arkose Labs / FunCaptcha: Enterprise-grade solving.

3. Smarter Navigation & Gates

• Email-Gate Bypass: Automatically completes "Submit email to enter" pre-forms (iCIMS/Breezy).
• SSO False-Positive Protection: Smart detection that only skips if no form exists, avoiding global navigation false alarms.
• LinkedIn Logic: Prioritizes direct ATS applications over LinkedIn Apply for higher success rates.

---

🧠 Multi-LLM Routing

HireAgent uses the right model for each task:

Scoring   →  NVIDIA DeepSeek-V3        (Cloud, high reasoning)
Tailoring →  NVIDIA DeepSeek-R1-14B    (Cloud, long-form writing)
Vision    →  NVIDIA Llama-3.2-11B-Vis  (Cloud, visual field mapping)
Mapping   →  NVIDIA Nemotron-340B      (Cloud, complex JSON schemas)
Enrich    →  Ollama gemma3:4b          (Local, fast extraction)

---

🛡️ Eligibility Filter — No Wasted Applications

Every job passes through a rule-based classifier:

• ✅ BS/MS roles (Entry-level / New Grad / Junior)
• ✅ US-based or Remote
• ❌ Senior / Staff / Principal / Lead
• ❌ Non-US location
• ❌ Requires Security Clearance

---

⚙️ Quick Start

1. Install Prerequisites

# macOS
brew install --cask mactex-no-gui   # LaTeX for PDF generation
brew install ollama && ollama pull gemma3:4b

2. Set Up API Keys

cat > ~/.hireagent/.env << EOF
NVIDIA_API_KEY=nvapi-...       # build.nvidia.com
CAPSOLVER_API_KEY=CAP-...      # capsolver.com
TELEGRAM_BOT_TOKEN=...         # For real-time unknown fields
TELEGRAM_CHAT_ID=...           # Notification destination
EOF

3. Run

hireagent doctor            # Verify setup
hireagent run               # Full pipeline: discover → tailor
hireagent apply --limit 5    # Start autonomous application

---

📋 Commands

hireagent apply                  # Apply continuously, score ≥ 7
hireagent apply --dry-run        # Fill forms but don't submit
hireagent apply --url <url>      # Apply to one specific job

---

📁 Project Structure

hireagent/
└── src/hireagent/
    ├── 🖥️  cli.py               ← All CLI commands
    ├── 📊 latex_renderer.py    ← LaTeX → PDF compiler
    │
    └── apply/
        ├── free_agent.py       ← Core state machine (SSO, Navigation)
        ├── vision_loop.py      ← Vision Fill + CAPTCHA Solver
        ├── launcher.py         ← Worker queue & Purge logic
        └── chrome.py           ← Playwright session manager

---

🙋 About the Author

Gunakarthik Naidu Lanka — MS Computer Science, Arizona State University (GPA 4.0)

---

📜 License

MIT © 2026 Gunakarthik Naidu Lanka — see LICENSE

_{⚡ Built with Python, Playwright, NVIDIA NIM, LaTeX, and way too many job applications.}