ApneaAlert

Inspiration

We have numerous family members who have been diagnosed with sleep apnea and struggle with airway blockage. Continuous positive airway pressure (CPAP) machines are designed to address the airway collapse associated with obstructive sleep apnea, which results from the relaxation of throat muscles. While instructed to use CPAP machines, however, they – alongside up to 83% of sleep apnea patients – are nonadherent, citing intrusiveness, discomfort, claustrophobia, and nasal congestion (Weaver and Grunstein). Furthermore, there are an estimated 936 million adults (ages 30-69) worldwide affected by sleep apnea (Iannella et al.).

When a patient with obstructive sleep apnea sleeps in the supine position (lying on their back), the tongue may fall back into the throat and lead to blockage of the airway. Such an episode can cause the patient difficulties breathing (observable as a snort, choke, or gasp), and this results in the brain repeatedly waking up to resume breathing – and simultaneously disrupting sleep (InformedHealth.org [Internet]; Mayo Clinic Staff). Furthermore, an extended inability to breathe could lead to adverse effects or death. While CPAP machines aim to address this issue, some patients still experience risks when in the supine position, and the risks are especially apparent for those who are nonadherent to the use of CPAP machines. To this extent, a study found that sleeping in the supine position was associated with significantly more apneas (lapses in breathing) than a patient in non-supine sleeping positions (Kavey et al.).

What it does

ApneaAlert’s primary purpose is to monitor the patient and alert them if they enter sleep in the supine position to reduce the risk of choking and/or apnea. The app utilizes a live video feed to detect sleeping position, designed to be compatible with external devices. Caretakers have access to this feed and receive instantaneous alerts in the case of the patient sleeping in the supine position. Furthermore, the patient and caretaker have access to a dashboard integrating sleep tracking and alert history to allow for informed next steps. Caregivers and patients are linked through unique patient codes to ensure patient safety. Patients can send notes to their respective caregiver, ensuring effective communication.

In regards to the user interface, our app is equipped with numerous features designed to enhance the user experience. Given the broad target audience of our app, accessibility is a key facet of our design. The user has the ability to adjust text size, toggle dyslexic-friendly font, reduce motion on the page, and increase contrast on page elements. Furthermore, users have the option to toggle between light and dark mode on the app.

How we built it

We worked with Gemini to finalize our project idea and developed our PRD in Claude. We used ChatGPT and Ace Logo Maker for app title and logos, respectively, followed by group voting on favored designs. Seven distinct designs were outlined in the Replit Canvas, from which we ultimately implemented the “Calm Clinical" format. While Replit and its associated Task Manager were used for much of our user interface (UI) development, we used Claude Code within the Replit shell in order to address the more complex functions in our backend. We integrated an ElevenLabs API for our voice-to-text feature for use in discussion with the AI health chatbot. Furthermore, model training was performed using the open-access IEEE VIP CUP 2021 Dataset available on Kaggle. Using the AMD Development GPU, we trained a total of six different models and selected the highest-performing of the following: ResNet-18, ResNet-34, ResNet-50, ResNet-101, ResNet-50 (from scratch), and ShuffleNet. Further model optimization was performed in Replit with the Tensorflow MoveNet. We used IP Camera Lite to allow for the connection of external devices for camera feeds.

Challenges we ran into

One challenge we ran into was implementing a live video feed from an external source, a key component of our project. We tested with numerous different tools before deciding to utilize the IP Camera Lite tool for current purposes. Another challenge we faced was in developing an effective model, especially given the limited data available to us. Ultimately, we tested multiple different models and performed optimization to achieve the accuracy that we demonstrate in our app.

Accomplishments that we're proud of

We are proud of developing a functioning app to address a real-world problem that we can use in our own lives and with our families. Adding to this, we are also proud of completing our first multi-day hackathon together!

What we learned

We learned how to utilize the ElevenLabs API for integration in our app. Furthermore, this was also our first time working with the MoveNet model from TensorFlow rather than the PoseNet model. In addition, we learned how to use the AMD GPU for model training and development. Topping it off, we learned how to integrate external devices for a streamlined live video feed in our app.

What's next for ApneaAlert

In regard to next steps for our ApneaAlert model, the model would benefit from a larger training database. We considered utilizing the “Simultaneously-collected multimodal Lying Pose (SLP)” from the Northeastern University ACLab for training, but ended up switching to the “IEEE VIP CUP 2021 Dataset” due to it being freely and instantaneously available on Kaggle. In the future, however, it would be beneficial to train on the significantly larger SLP Dataset (14,715 images) to improve model efficacy.

In terms of the app itself, the next step for ApneaAlert would be facilitating and testing direct connection with tools, including but not limited to Ring cameras, which are equipped with night vision, to allow for an effective video stream. Furthermore, we would also perform app security testing using the SecureVibes multi-agent tool for vulnerability identification, available on GitHub.

Citations

Iannella, Giannicola, et al. "The global burden of obstructive sleep apnea." Diagnostics 15.9 (2025): 1088. InformedHealth.org [Internet]. Cologne, Germany: Institute for Quality and Efficiency in Health Care (IQWiG); 2006-. Obstructive sleep apnea: Learn More – Treatment of obstructive sleep apnea. [Updated 2022 Dec 19]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK279271/ Kavey, Neil B., et al. "Sleeping position and sleep apnea syndrome." American journal of otolaryngology 6.5 (1985): 373-377. Mayo Clinic Staff, “Sleep Apnea”, Mayo Foundation for Medical Education and Research, 9 Dec. 2025, www.mayoclinic.org/diseases-conditions/sleep-apnea/symptoms-causes/syc-20377631. Weaver, Terri E., and Ronald R. Grunstein. "Adherence to continuous positive airway pressure therapy: the challenge to effective treatment." Proceedings of the American Thoracic Society 5.2 (2008): 173-178.