Inspiration The inspiration for KeepSchoolsSafe.ai stems from a sobering reality: since 2018, hundreds of school shootings resulting in injuries or deaths have occurred on K–12 campuses across the United States. Active-shooter drills have tragically shifted from rare precautions to routine, everyday occurrences for students and educators. Disturbed by this ongoing safety crisis, we wanted to build a proactive solution. Instead of waiting for an incident to occur, we envisioned a system that could act as an immediate line of defense to protect innocent lives.
What it does KeepSchoolsSafe.ai is an advanced security platform that uses artificial intelligence and computer vision to turn standard school security cameras into intelligent threat detectors. The system scans live camera feeds in real time to identify brandished firearms or potential weapon threats. The moment a threat is verified, the platform bypasses delayed human reporting to instantly dispatch high-priority alerts to school administrators, resource officers, and emergency personnel—cutting down critical response times when every second counts.
How we built it We engineered the platform using a layered modern tech stack:
Computer Vision & AI: We trained a custom object-detection model (such as YOLO) on a robust dataset of firearms and tactical objects to optimize for low-latency, real-time edge processing.
Backend Infrastructure: Built using Python (FastAPI/Flask) to handle live video stream processing and route webhooks instantly.
Frontend Dashboard: A clean, intuitive dashboard developed with React that visualizes camera feeds, maps out threat locations on a school blueprint, and highlights real-time alerts.
Notification Engine: Integrated with communication APIs (like Twilio) to immediately send automated SMS and push notifications containing the exact threat location and a video snippet to first responders.
Challenges we ran into Our biggest hurdle was striking the perfect balance between speed and accuracy. In high-stakes security, false positives (like a student holding a broom or a long umbrella) can cause unnecessary panic, while a false negative is unacceptable. Fine-tuning the confidence thresholds of our computer vision model to handle poor lighting, motion blur, and low-resolution camera feeds required rigorous testing. Additionally, managing heavy data loads from multiple simultaneous live video streams without causing server lag was a significant architecture challenge.
Accomplishments that we're proud of We are incredibly proud to have built a functional, end-to-end prototype that can detect a weapon in a live stream in under a second. Successfully integrating a real-time alerting system that maps the exact camera location on a school floor plan feels like a major victory. Knowing we built a tool that handles a real-world, high-stakes issue with practical, life-saving potential is our proudest achievement.
What we learned This project taught us an immense amount about the complexities of deploying real-time AI on video streams. We learned the nuances of dataset curation, specifically how critical it is to include "negative" examples (like phones, tools, and toys) to teach the model what not to flag. On a human level, researching school safety data deepened our empathy for the educators and students living through this crisis, reinforcing the concept that technology should always serve human well-being first.
What's next for KeepSchoolsSafe.AI Moving forward, we want to expand the platform's capabilities beyond visual detection:
Audio Integration: Adding acoustic sensors to detect and cross-verify gunshots or aggressive audio patterns.
Predictive Anomaly Detection: Implementing behavior analysis to flag suspicious activity, such as forced entries or tailgating at restricted access points.
First Responder Integration: Collaborating directly with local emergency dispatch services to automatically feed live camera links directly to responding police units, giving them eyes inside the building before they arrive.
Log in or sign up for Devpost to join the conversation.