auto-apply-agent 🤖 — An AI agent that fills job applications for you.

An autonomous AI agent that fills out Greenhouse job applications end-to-end. Navigates multi-step forms, answers screening questions using context from my resume, handles file uploads, and manages security codes — all without human intervention.

This is a closed-source project. The README documents the architecture and learnings.

What it does

• Fills Greenhouse applications autonomously from URL to submission
• Answers free-text and multiple-choice screening questions with LLM
• Uploads resume and cover letter (generated per-application)
• Handles CAPTCHA and security code challenges
• Manages browser sessions with automatic cleanup and memory leak prevention
• Detects and recovers from stale page states

How it works

Job URL ──► Launch Browser ──► Navigate to Form
                                     │
                                     ▼
                              Map Form Fields
                                     │
                    ┌────────────────┼────────────────┐
                    ▼                ▼                ▼
              Fill Personal    Answer Questions   Generate Cover
              Info (profile)   (Claude + resume)  Letter (Claude)
                    │                │                │
                    └────────────────┼────────────────┘
                                     ▼
                              Upload Documents
                                     │
                                     ▼
                              Handle Security
                              Code (email poll)
                                     │
                                     ▼
                                  Submit
                                     │
                              ┌──────┴──────┐
                              ▼              ▼
                           Success       Auto-Restart
                                        (crash recovery)

Error recovery includes auto-restart on crash, session cleanup, and DOM polling with exponential backoff.

Tech stack

• Runtime: Node.js with Puppeteer for browser automation
• AI: Claude via Bedrock for screening question answers and cover letter generation
• Browser: Headless Chromium with stealth plugins
• Infra: AWS Lightsail, PM2 for process management

What I learned

• Browser automation at scale requires treating the browser as an unreliable service — expect crashes, leaked memory, and stale sessions. Every run needs a cleanup guarantee.
• LLMs are surprisingly good at answering "Why do you want to work at X?" when given the job description and your background. The output is better than most human rush-job answers.
• Security code handling was the hardest edge case — you need email polling with timeout and retry, and the code expires fast.