ExecuTorch Hackathon winners show on-device AI for safety, accessibility

The ExecuTorch Hackathon, held June 27–28, 2026, in San Francisco, gathered over 100 developers from 20+ teams to build real-time on-device AI applications using PyTorch and ExecuTorch on Snapdragon-powered Samsung Galaxy S25 Ultra phones. Supported by Meta, Qualcomm, the PyTorch Foundation, and GitHub, the event focused on latency, offline capability, privacy, and energy efficiency. Winning teams demonstrated practical edge AI: SafeScreen AI took first place with an on-device visual safety layer that detects and blurs explicit or manipulated media in real time, keeping processing local for privacy. Second place went to SixthSense, a haptic assistive system that uses phone cameras to detect obstacles and sends directional vibration signals to a waist belt, helping blind users navigate without cloud connectivity.

The hackathon underscored a growing shift from cloud-only AI to edge deployment for real-time, privacy-sensitive applications. ExecuTorch, an end-to-end runtime from the PyTorch Edge ecosystem, enables efficient inference on CPUs, NPUs, and DSPs across mobile, wearables, and microcontrollers. Projects ranged from accessibility tools and privacy-first assistants to offline document intelligence and medical support. The event highlighted how on-device AI can deliver responsive, cost-effective experiences where connectivity and data privacy are critical, setting the stage for broader adoption in professional and consumer contexts.