AI-Powered Malware Detection
Real-time process surveillance that watches resource behaviour across running processes and flags anomalies with a Bidirectional LSTM.
The problem
Signature-based detection only catches malware someone has already seen. Behaviour is harder to disguise than a file hash.
How it works
The system continuously samples CPU, memory, and resource usage for every running process, turning each into a time series of behavioural features rather than a static snapshot.
A Bidirectional LSTM trained on those sequences scores processes for anomalous behaviour, with a threshold-based heuristic running underneath it as a fallback layer — so the tool still flags obvious resource abuse when the model is uncertain.
A PyQt5 desktop UI surfaces live per-process risk, because a detector nobody watches is not a detector.
Outcomes
- Runners-Up at HackTheThreat, Abhisarga '25.
- Layered detection: learned model plus deterministic heuristic fallback.
- Built with a teammate over a hackathon weekend.