Seeking Feedback: Real-Time Screen + Keystroke Monitoring for AI-Aware Anti-Cheating System (FYP)
🧠 Title:
Seeking Feedback: Real-Time Screen + Keystroke Monitoring for AI-Aware Anti-Cheating System (MVP FYP Project)
📝 Post Body:
I’m a CS undergrad working on my Final Year Project, and I’d really appreciate some constructive critique from the developer, ML, and privacy-conscious communities.
🔍 Problem:
With remote learning and online exams becoming common, academic dishonesty is increasingly hard to detect — especially with the rise of LLMs, copy-paste coding, and browser switching during assessments.
Current proctoring tools focus mostly on webcams and raise serious privacy concerns, while still being easy to bypass.
💡 Our MVP Proposal:
We're building a real-time, privacy-conscious anti-cheating system focused on:
Live screen stream monitoring (1–2 FPS sampling for efficiency)
Real-time keystroke analysis (flagging ctrl+c, ctrl+v, AI keywords like "ChatGPT", etc.)
Tamper detection (VM detection, sandbox evasion, plugin/modification flags)
Automated flagging via lightweight ML — only shows partial logs that triggered the alert
Auto self-destruct after the exam to eliminate data persistence or tracking concerns
We’re deliberately not using webcams, microphones, or storing full keylogs/screens. Only flagged behavior is logged.
🔐 Privacy Policy Safeguards:
App runs only during exam, self-uninstalls afterward
No webcam/audio access, no biometric tracking
Students agree via EULA + pre-exam consent
Source code will be partially open for transparency
🧪 Architecture (Draft)
Frontend: Electron-based cross-platform exam app
Monitoring Layer: Native C++/Rust agent for screen & process monitoring
Backend: Python API with flag logic, hosted on secure VPS (10–1000 concurrent streams)
ML: Lightweight detection models for anomaly + AI usage flags (not deep surveillance)
💬 My Ask:
Is this technically viable at scale (1K students)?
What are the most critical flaws in this design?
How can I maintain control without violating ethical boundaries?
Would you (as a developer or educator) trust a system like this?
🙏 Why This Matters:
If we can strike the right balance between cheating detection and privacy protection, we might be able to offer a legitimate solution to universities struggling with online examination integrity — without turning every student's room into a surveillance state.
All feedback — critical or supportive — is welcome.
Thanks in advance.
Replies
Building an anti-cheating system that tries to respect privacy is a noble goal, but from an architecture standpoint, you’re walking a massive tightrope.
The biggest technical flaw in the MVP design is the C++/Rust native agent combined with Electron. The moment you require students to install a native executable that hooks into process monitoring, screen capture, and keystroke logging, university IT departments and security-conscious students are going to panic—regardless of your self-destruct promises. Antivirus software will instantly flag it as spyware (trojan/keylogger signatures), and getting past Windows Defender or macOS Gatekeeper for unverified student builds is a nightmare.
If I were evaluating this as a scalable system:
The 1K concurrent stream load: Streaming screens and processing keystrokes via a Python backend on a standard VPS will bottleneck fast. You’ll need a robust message broker like Redis or RabbitMQ and a pub/sub architecture to handle real-time anomaly detection without crashing the API.
The bypass problem: Students will simply use a second device (phone/tablet) placed next to the monitor. Keystroke analysis and screen monitoring won't catch someone looking at ChatGPT on their phone.
To answer your trust question: as a developer, I wouldn't trust a custom native proctoring agent on my personal machine. If universities want secure testing, they usually push for managed lockdown browsers (like Safe Exam Browser) rather than custom-built telemetry agents.
How are you planning to handle the false-positive nightmare on the keystroke side? A student rapidly hitting Ctrl+C and Ctrl+V while refactoring their own code in an IDE is going to trigger every alert you have.