-
CTRL+ALT+DROPOUT
Inspiration
The challenge presented a complex multi-step workflow with multiple CAPTCHAs, payment processing, and security layers, the perfect testbed for an autonomous AI agent. Plus, who hasn't fantasized about automating the dropout process at 3 AM during finals week?
What it does
YOLO: You Only Logout Once is an autonomous web automation agent that completes the entire university student dropout process from start to finish:
- Authenticates through login with netname/password
- Defeats 3 types of CAPTCHAs:
- Logo recognition using computer vision (OpenCV)
- Numeric sequence puzzles (click numbers in order)
- Handles Two-Factor Authentication by extracting MFA codes from popup windows
- Processes payments
- Navigating popup windows
- Selecting saved payment methods
- Solving additional CAPTCHAs in the payment flow
- Drops all enrolled courses automatically
- Completes the dropout process
All without human intervention. Speedrun time: ~3 minutes ⚡
How I built it
Tech Stack:
- Python - Core automation logic
- Playwright - Browser automation and web scraping
- OpenCV - Computer vision for logo CAPTCHA detection
- NumPy - Image processing and analysis
- Regex - Pattern matching for MFA codes and sequences
Key Technical Challenges:
- Logo CAPTCHA Solver: Used HSV color space analysis and edge detection to identify logos vs solid colors
- Sun CAPTCHA Trick: Had to detect the inverse (I waisted a lot of time figuring it out haha)clicking images that are NOT suns required yellow/orange ratio analysis with a threshold
- Popup Window Handling: Payment processing opened in a new window context requiring proper Playwright popup management
- Dynamic Scrolling: Multiple instances where content wasn't visible without programmatic scrolling
- State Management: Tracking workflow progress across 9+ distinct steps
Challenges I ran into
- The Sun CAPTCHA Betrayal: Spent hours perfecting sun detection with HSV color analysis only to discover the instructions wanted us to click NON-sun images. Classic misdirection! 🌞
- Popup Hell: The payment flow opened in a popup window, which required completely different context handling in Playwright
- CAPTCHA Evolution: Each CAPTCHA type required a different solving strategy (regex for numeric sequences, computer vision for logos, color analysis for images)
- Timing Issues: Finding the perfect balance of wait times between actions to ensure page loads completed without wasting time
- Scroll Mechanics: Buttons and forms were often off-screen, requiring strategic scrolling at precise moments
Accomplishments that I am proud of
Built a fully autonomous agent that completes a 20+ step workflow without human intervention
Defeated 3 different CAPTCHA types using custom algorithms
Successfully processed payment in a popup window (harder than it sounds!)
Computer vision implementation that actually works in production
The agent has a 100% success rate when CAPTCHAs are solvable
Created something genuinely useful (and slightly chaotic)
What I learned
- Computer vision is simultaneously easier and harder than expected
- Web automation requires thinking like the website instead of fighting it
- Sometimes the best solution is inverting your logic (looking at you, sun CAPTCHA)
- Popup windows are the final boss of web scraping
- Error handling and screenshots are your best friends when debugging
- The University bureaucracy can actually be beaten with code
What's next for YOLO: You Only Logout Once
- Generalize the agent to work on other university portals
- Add machine learning to improve CAPTCHA solving accuracy
- Create a GUI for non-technical users
- Support for more CAPTCHA types (reCAPTCHA v3, hCaptcha, etc.)
- Parallel processing for handling multiple accounts
- Docker containerization for easy deployment
- Maybe build the opposite: an automated enrollment agent?
Log in or sign up for Devpost to join the conversation.