RAKSHAK: A Multimodal Android Safety Platform Integrating Discreet Emergency Triggers and Responsible AI-Driven Support

Abstract

Women’s safety apps have traditionally employed conventional panic button designs, which are immediately rendered ineffective as soon as the user finds herself in a situation where she cannot openly touch her screen. This paper proposes RAKSHAK, a native Android platform based on two unconventional, low-probability emergency triggering techniques—a computer vision pipeline for detecting deliberate eye blink patterns and an accelerometer-based signal processing pipeline for detecting deliberate triple tap rhythms. Along with these low-probability emergency triggering techniques, RAKSHAK also offers proactive assistance in the form of live sharing, a facility-finding mechanism based on proximity, informative content, and a scope-bound AI-powered chat assistant based on the Gemini large language model. The efficacy of RAKSHAK was validated through a study involving twenty participants, with results indicating a range of 85.5% to 99% for the true positive rate of the emergency triggering techniques and a SUS score of 82.5. An ethical architecture for the scope-bound AI-powered chat assistant offered in RAKSHAK serves as a reusable template for the safe deployment of generative AI in safety-critical mobile applications.