AlignAI

Inspiration

This app was inspired by our everyday struggle when working or studying for long hours: maintaining good posture. While we know having good posture is extremely important, continuous self-awareness of it can be difficult.

We recognize that existing posture tools may be more rigid or clinical, which focus on deficits rather than support. By combining cutting-edge AI posture tracking with a system of friendly encouragement, personalized feedback, and fun rewards, we aimed to build a solution that's interactive, supportive, and actually approachable.

What it does

AI Posture Detection

Uses your camera feed to monitor posture in real time
Detects when you’re sitting upright vs. slouching
Categorizes your type of posture

Friendly Posture Buddy

Animated buddy reacting to your posture.
Gives gentle reminders of posture status in real time

Session Tracking

Start and end posture monitoring session
Tracks total time and good posture time

Smart Analysis & Feedback

End-of-session breakdown with charts to show posture statistics
AI-generated feedback + personalized posture improvement tips.

How we built it

Our web application was created with a Python/Flask backend for AI processing and a standard HTML/CSS/JavaScript frontend for the user interface.

Major Components

MediaPipe library for posture detection
Gemini API for generating the final report
SQLite for data persistence

Challenges we ran into

Calibration (Personalization): We had to figure out what was an ideal upright position for each user, which required the use of a personalized calibration method based on each user's sitting position. This involved developing a method to capture baseline angles during a brief initial calibration phase, ensuring the system was effective across various body types and chair setups.
Handling False Negatives/Positives: We struggled with achieving a good threshold for the angle movement that resulted in negative posture values. This required iterative data collection and fine-tuning of threshold angles to prevent the system from giving constant, false warnings on incorrect positions.
Integration of AI Summarization: Integrating the Gemini API for contextual feedback presented a challenge. We had to design the data pipeline and prompts to efficiently collect relevant session data without exceeding token limits while ensuring the output was still insightful and helpful.