TBMate
TBMate is a mobile health platform designed to help Puskesmas monitor tuberculosis patients' medication adherence remotely, especially in rural areas where regular in-person visits can be difficult. The app allows patients to verify their medication intake from home, while healthcare workers gain better visibility into treatment consistency.
Project Type
Group Project
Project For
Puskesmas Kabat, Banyuwangi, Jawa Timur
Role
AI Engineer
Platform
Android
Year
2026
Technology Stack
The Challenges & Solutions
Optimizing for Low-End Devices
Our early prototype used real-time video analysis, but testing showed that it caused **lag, overheating, and poor performance** on lower-end smartphones. Instead of forcing a technically impressive solution, I redesigned the flow into a sequential photo-based verification system using Google ML Kit. This made the experience lighter, more stable, and more realistic for patients in rural environments.
Ensuring Pill Verification Safety
Because tuberculosis treatment depends on consistent and accurate medication intake, the verification process needed to minimize the risk of **false confirmation**. I developed a computer vision flow that analyzes captured pill images and applies confidence thresholds before recording the medication status. This helped make the system more careful, especially in a health-related context where incorrect verification could affect patient monitoring.
The Impact I Made
By designing a **hardware-friendly AI verification flow**, I helped move the prototype closer to real-world implementation with Puskesmas Kabat by aligning the solution with healthcare worker needs and patient device limitations. I improved app performance by replacing heavy real-time video analysis with a **lighter photo-based verification flow** suitable for lower-end devices, supported healthcare workers with a more practical monitoring flow, and created a safer verification process by reaching **98% precision in internal prototype testing** under controlled image conditions.
What I Learned
TBMate taught me that the best technical solution is not always the **most advanced one**. In real healthcare environments, reliability, accessibility, and patient comfort matter as much as model performance. By pivoting from real-time video AI to a lighter photo-based flow, I learned to design around real constraints such as device limitations, internet stability, and user effort. This project helped me understand that **meaningful AI should adapt to people's reality**, not force people to adapt to the technology.