Guardian Angel

Guardian Angel

Guardian Angel is an AI-powered mass casualty perception system designed to augment combat medics during MASCAL events in Tactical Combat Casualty Care (TCCC) operations. Built for disconnected, intermittent, and low-bandwidth (DIL) tactical edge environments, Guardian Angel multiplies a medic's perceptual capacity across an entire casualty scene simultaneously — replacing sequential, manual triage with real-time AI-assisted situational awareness. Every minute a victim goes without rescue, they chances of survival decrease by 6%. Guardian Angel saves lives by increasing the speed of search and rescue, pushing for rescue with the golden hour. Drones and people can transmit geotagged pictures from the battlefield which are rapidly assessed by AI. Medics can share geotagged voice notes hands free without typing. The system consists of three securely integrated components. The AI Workflow processes visual imagery from drones or mobile cameras using Meta's SAM 3 segmentation model to detect, segment, and classify casualties according to the SALT triage protocol — assigning doctrine-standard categories of Immediate, Delayed, Minimal, and Expectant in real time. The Voice App enables combat medics to verbally report casualty status using a mobile interface, transcribing speech offline using Faster-Whisper before transmitting to the command console. The Guardian Angel Console provides incident commanders with a single unified view of the entire casualty scene, including a live SALT status board, evacuation priority queue, transmission log, and medic notes — updated in real time as reports arrive from the field. Casualty data flows forward to the next role of care and back up to the command node for resource allocation and MEDEVAC coordination — compressing the time between point of injury and life-saving intervention. Guardian Angel is grounded in TCCC doctrine, validated against the SALT protocol, and designed for the conditions combat medics actually operate in — gloves on, under fire, with no guarantee of air evacuation or cloud connectivity. The code was written with the help of OpenAI codex which automatically reduces risk by writing functions that adhere to best code security practices. Equally important, OpenAI models synthesized test data input, such as mascal pictures and mp3 medic voice notes. This version is non-exhaustive, lean, and runs on regular hardware. An improved version would involve hosting on headset wearables for fast picture taking. Secure login would control access to the console. Drones with cameras would be equiped with software to transmit pictures to Guardian Angel APIs. High powered GPUs hosted by model would speed up SAM3 inference. Finally, additional crypto would further secure the data in transport and in rest.