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.