AI-Cyber-Guardian Public

🚀 Inspiration The internet is a vast place, but it’s not always safe. Traditional filters rely on "static lists" of bad words. If a word isn't on the list, the filter fails. I wanted to build something smarter. I was inspired by the idea of a digital butler—like Jarvis—that doesn't just look at words, but understands intent. I wanted to create a guardian for my own system that could think like a human but act with the speed of an AI.

🧠 How I Built It I used Python as the backbone of the project. The "eyes" of Jarvis are powered by the pygetwindow library, which monitors active window titles in real-time. The "brain," however, is the Gemini 1.5 Flash API.

Every time a window title changes, the text is sent to Gemini with a specific system prompt.

The AI analyzes the semantics and context of the title rather than just searching for keywords.

If the AI returns a DANGER verdict, the system executes a lockdown protocol using Windows system calls.

🛠 Challenges I Faced The biggest challenge was the "Escape Artist" problem. Initially, the program could be closed simply by clicking the "X" button on the console. To solve this, I implemented a Ghost Process logic using ctypes to hide the console window entirely, making the protector invisible and un-closable by standard means. Another challenge was "False Positives." I had to refine my Prompt Engineering to ensure that educational content stays accessible while actual inappropriate content is blocked instantly.

📖 What I Learned During this Hackathon, I learned:

Prompt Engineering: How to force an LLM to give binary (DANGER/SAFE) outputs with 100% reliability.

Windows API Integration: Managing system-level tasks like locking the workstation and killing specific browser processes.

Real-time AI Processing: Balancing the frequency of API calls with system performance to ensure zero lag.

Built With

• python