Developer Builds AI-Powered Scam Detector for Family Using Open-Weight Gemma

A developer has created 'Chithi,' an open-source tool designed to protect non-English speaking family members from financial scams. The application uses Google's open-weight Gemma models to analyze photos of documents, bills, and SMS messages, identifying potential threats in 22 Indian languages. By running the model locally on a home laptop via Ollama, the developer ensures that sensitive personal data, such as bank letters and medical prescriptions, remains private and is never sent to third-party servers. The system features a deterministic rule-based layer that prevents the AI from providing dangerous advice, such as calling scammer-controlled phone numbers. The project emphasizes the importance of local, open-source AI for accessibility and privacy, allowing users to maintain control over their security without relying on subscription-based services. The tool is available on GitHub and demonstrates a practical, privacy-focused application of vision-language models for vulnerable populations.