ApexShield

Two AI-powered apps: one delivers free, personalized self-defense training from the USA Taekwondo National Team athlete, the other enhances safety for the visually impaired using Raspberry Pi vision.

Comment

note: the ApexVision Demo is linked at the top, but please check out the ApexDojo Demo as well, in the Try it Out section. Thank you!

Inspiration

What inspired me to create this project was witnessing unsafe events where people didn't have the ability to protect themselves. In life-threatening situations like that, it's difficult to get quick help. People with disabilities like visual blindness are even more disadvantaged, and I wanted to build technology that protects them in their daily-life. I wanted to use artificial intelligence to make safety more accessible. As someone who's been doing taekwondo for 14 years, competing in the World Class Sparring Black Belt Division, being a 3X USA National Team Member, 11X National Champion and representing and winning Pan-American Championships, I've always wanted to share the knowledge I've acquired in a way that's free, accessible, and easy to understand for all. That led me to build two connected applications: ApexShield Dojo, which provides free AI-powered self-defense training for everyone, and ApexShield Vision, which helps visually impaired individuals better understand their surroundings through real-time object detection and voice guidance.

What it does

The first app, ApexShield Dojo has 4 main pages. The first is the AI Poomsae Pose Coach. This contains many blocks, stances, and kicks that users can practice. Users on their cameras and practice the move the AI teaches. Then, it uses pose estimation technology to analyze the motions of the athlete and gives live feedback, with a score, and what exactly to improve to get it perfect. The second page is the AI Poomsae Library, which has videos of me and tutorials by AI that teach all Taekwondo Poomsaes. The third page is the Full Poomsae Analyzer, where users can record themselves doing a full poomsae, and it will analyze the frames afterwards using Media Pipe technology and give an accurate score with personalized feedback. The last page is the Self-Defense tutorials, where I recorded videos teaching the real-world applications of taekwondo moves when threatening situations arise. The second app, ApexShield Dojo's main application is the object detection for safety. Using a Raspberry Pi, camera, computer vision, and AI object detection, the system identifies nearby objects and potential hazards, then communicates important information through voice and haptic feedback. For example, if a stranger approaches with an object in their hand, the camera will quickly detect this and notify the user that "Person quickly approaching with threatening object in hand. High alert." Then, the user also has a voice feature where if they speak the code word in an emergency situation, a loud siren will play and they can call 911, trusted contact, and automatically on their GPS. The goal is to increase situational awareness and help users navigate everyday environments more safely.

How we built it

I developed both apps initially using Bolt, through Python code. Then, to make better updates, I used Kiro IDE to make personalized updates to a local host, so many versions were available. I deployed the web application through Vercel. The mobile application was converted into a native iOS app using Capacitor and Xcode. For ApexShield Dojo, I integrated AI pose estimation to analyze body landmarks captured by the camera and evaluate Taekwondo techniques in real time, generating personalized feedback and scoring. ApexShield Vision combines a Raspberry Pi camera, computer vision, and AI object detection models like COCO-SSD to identify nearby objects and communicate them through voice guidance. I also integrated Google Analytics to understand user engagement and continuously improve the experience based on real usage. Small Snippet of the Code {screen === 'poomsae-detail' && ( <motion.div key="poomsae-detail" variants={pageVariants} initial="initial" animate="animate" exit="exit" transition={transition}> <PoomsaeDetailScreen poomsaeId={selectedPoomsaeId} onBack={() => setScreen('poomsae-library')} onPractice={(skillId) => { setPracticeSkillId(skillId); setScreen('taekwondo'); }} /> </motion.div> )}

Challenges we ran into

Some challenges I ran into was making real-time pose estimation accurate enough to provide meaningful coaching across different users, camera positions, and lighting conditions. I also encountered difficulties with making the object detection in the vision app accurate. Often times, it would misinterpret the object a person was holding if they were coming fast, but with code on Kiro, I was able to make it smart by giving specific situations to go off of.

Accomplishments that we're proud of

I'm really proud of the emergency feature enabled through voice on the Vision App. I think this addition really helps seal the safety aspect of the app as users may not have time or the state of mind to escape immediately. Instead of clicking buttons, they can just say the code word and the mode gets activated. I'm also proud of using a small raspberry pi and camera on a cap, so users blend in and they're able to use the prototype easily. I'm also really proud that the Vision app works through the phone hotspot, so internet access isn't needed. I'm also proud that I made professional self-defense education freely accessible through interactive AI coaching.

I'm also proud of the Apex Dojo App reaching 100+ users and athletes and coaches giving testimonials on the app. It really heartens and motivates me to know that I'm making a real impact on people's lives in a subject that is often overlooked. I feel self-defense is so important for everyone and it's not just a skill, but a way of life.

What we learned

This project taught me that building an AI model is only one part of creating a useful product. Delivering reliable real-time performance, designing an intuitive user experience, integrating hardware with software, and deploying applications across multiple platforms presented equally important engineering challenges. I also gained experience working with computer vision, pose estimation, mobile development, cloud deployment, and iterative product development driven by user feedback.

What's next for ApexShield

So far, I've gotten over 345 events by 131+ unique users for the ApexShield Dojo App and testimonials from accredited users who benefitted from my app. is to expand ApexShield into a comprehensive AI-powered personal safety platform. For ApexShield Dojo, I plan to support additional martial arts styles, improve pose estimation accuracy, introduce adaptive training plans, and expand the AI coaching system. For ApexShield Vision, I aim to enhance offline functionality, improve object recognition accuracy, integrate emergency voice commands, and continue testing with visually impaired users to better understand their needs. Long term, I hope to make advanced personal safety technology freely accessible to anyone who needs it. I want to continue expanding both apps and potentially partner with martial arts companies as well as Meta or other AI companies. I want to continue to improve the accuracy and functionality of both apps.

Built With

Share this project:

Updates

Submission history