Inspiration
Have you ever downloaded productivity or blocker apps to stop yourself from scrolling or overusing certain apps only to uninstall them a few days later? The core problem with these tools is that there are no real consequences for turning them off and giving in to procrastination. Without accountability, it’s easy to ignore the rules you set for yourself. We wanted to change that by introducing a real, immediate consequence for getting distracted: getting sprayed by a water gun.
What it does
Wet Reminder is a web application that helps users stay focused while studying. Users can set study timers and break intervals, while an LSTM-based deep learning model runs in real time to classify whether the user is focused or distracted. When the model detects distraction, it sends a signal to an Arduino-controlled motor that activates a water gun, snapping the user back into focus and discouraging further procrastination.
How we built it
We built the system by collecting and labeling our own training and testing data. Using the Mediapipe library, we extracted keypoints from face meshes, hand meshes, and other landmarks. These keypoints were then fed into an LSTM model trained on sequences of 150 frames (approximately 5 seconds of video per clip). The trained model runs in real time alongside the web interface and communicates with the Arduino hardware when distraction is detected.
Challenges we ran into
- Displaying face and hand meshes correctly on the frontend while running the model simultaneously.
- Tuning the LSTM architecture and selecting appropriate parameters for optimal performance.
- Installing Python
- Stabilizing the water gun, servo and the arduino with the mount to generate enough force for a good squirt of water.
- Google CoLab not being able access OpenCV and have access to our web-cam for real time testing.
- 26,000 untracked files and 3 separate .venv files.
Accomplishments that we're proud of
- Achieved a solid 85% level of accuracy with the distraction detection model
- Built a smooth, responsive user interface that works reliably across different webcams
What's next for Wet Reminder
We plan to expand the model to detect additional types of distractions, such as sleeping, gaming, or other off-task behaviors. We also want to significantly increase our training dataset (currently limited to around 60 videos) to improve robustness and accuracy across more scenarios.
We also plan to update the GUI and add more features that allow for a more cohesive user experience, such as feature selection to adapt to habits and summaries of the data tracked across sessions.
Log in or sign up for Devpost to join the conversation.