Solution = ATMAN — Adaptive Threat & Mindfulness Awareness Nudge ATMAN is an on-device(locally running) wellness companion that runs silently in the background of your Android device(Android 9 (API 29) minimum), intelligently aware of what the user is doing(samples the active foreground application periodically to infer the user's current activity category — all processing happens on-device). Using an on-device LLM (Google's Gemma model/Phi-2), it generates real-time, context-sensitive security nudges delivered as notifications — no internet required, no data leaving the device. What makes ATMAN different is its timing. Traditional training delivers awareness in a classroom or a simulated drill — long before or after the moment of actual risk. ATMAN delivers it at the moment of exposure, when the user is most receptive and the insight is most actionable. It overcomes three core failures of traditional approaches: Passive → Active: Awareness is triggered by real behavior, not a scheduled test Generic → Contextual: Every nudge is generated fresh by an LLM based on what the user is actually doing Forgotten → Reinforced: Repeated, lightweight nudges build lasting habit rather than one-time recall ***Privacy is preserved by design — all inference happens on-device, meaning sensitive usage data never leaves the user's phone, no internet is required for running. Hence can be trusted by normal users.

Built With

  • and-apis-from-hugging-face-for-inference-and-model-usage.-used-cloud-based-training/deployment-via-openenv
  • android
  • combining-ai/ml-pipelines-with-mobile-frontend
  • data
  • integrated-with-mediapipe-pipelines
  • javascript-(react/react-native)-for-android-app-development
  • llm-models
  • real-time
  • rest-apis
Share this project:

Updates

Submission history