ScamStop was inspired by a personal experience—my grandmother fell victim to a phone scam by fraudsters impersonating a utility company. This made us realize how vulnerable people, especially elders, are to these deceptive tactics. We wanted to build a tool that could intervene in real time, preventing these scams before any damage is done.
ScamStop listens to ongoing conversations and analyzes them in real time to detect scam patterns. Using AI-powered language processing, it identifies warning signs such as urgency, threats, or requests for sensitive information. If a scam is detected, the system issues an alert, helping users recognize fraudulent calls before they fall victim.
Frontend: HTML/CSS with vanilla JavaScript Backend: Flask server managing audio processing and ML model interactions Real-time Speech-to-Text: RealtimeSTT for continuous transcription AI Model: Ollama running locally with olmo2:13b for scam detection (can use different models) Process Management: Python subprocess handling for continuous recording and analysis
Managing microphone resources between processes Implementing real-time updates without overwhelming the system Coordinating multiple processes (recording, transcription, and detection) Ensuring reliable communication between frontend and backend components Balancing speed and accuracy in scam detection RELIABILITY
Implemented continuous speech-to-text processing Built a web interface for monitoring Developed a system that can help protect vulnerable populations Successfully integrated local LLM for privacy-conscious analysis
Real-time audio processing techniques Process management in Python Frontend-backend communication patterns Local LLM integration
Improve detection accuracy with our own model Add support for multiple languages Implement call recording for evidence Create mobile applications for wider accessibility (privacy features may make this hard) Add more sophisticated scam pattern recognition