WHAT I BUILD
This is my build log: the things I've made, broken, and made again. The ones I care about most started with a single question: how can a family know if someone they love is breathing easier today than yesterday?
AI Lung Care
THE PROBLEM: For lung cancer patients, especially older ones, progress is measured in scattered pieces: a nodule size from one CT scan, a note about coughing from another week, a doctor's comment nobody wrote down. Families like mine end up holding the whole picture in their heads.
WHAT I BUILT: Lung Progress Tracker lets patients and caregivers log nodule measurements, tumor growth rate, and the frequency of respiratory symptoms, then shows them as clear timelines. Instead of "I think she's coughing more," a family can see it — and bring that chart to the next appointment.
HOW IT WORKS: I designed the data model and visualization layer, and led a team of five students interested in AI and healthcare. We tested the app on 12+ cases over more than 90 days. Doctors at K Hospital reviewed our data handling and gave feedback on what clinicians actually need to see, which changed our design more than any line of code.
THE RESULTS: We brought the app to five community
health stations in Hòa Bình and Pù Luông. There I learned something no dataset had taught me: many elderly patients don't own smartphones, and many clinics have no CT scanner nearby. A tracker only works if there is something to track. That lesson became my next project.
Developer & Team Lead (05/2024 - Present)
Lungcare Pi Camera
Inventor (12/2025 - Present)
The problem: Detecting abnormalities in lung CT scans isn't always straightforward. Small, suspicious regions can be difficult to recognize, especially for those without specialized medical training. I wanted to explore whether AI could make these abnormalities easier to identify and examine, using hardware that is both affordable and accessible.
What I built: LungCare Pi Camera is an AI-powered prototype that uses a Raspberry Pi 4 and a camera module to analyze lung CT images. Instead of requiring powerful computers or specialized hardware, the system runs an AI model directly on a compact device, highlighting potentially abnormal regions with bounding boxes and confidence scores.
How it works: An OV camera captures CT images from a screen or printed scan. The system converts each frame into grayscale, isolates the relevant area, and feeds it into the LungCare model to identify suspicious regions. When the model detects a potential abnormality, it highlights the area on screen. Users can also adjust the detection threshold, zoom in, and examine the results in real time. Everything runs on the Raspberry Pi's CPU, without a dedicated AI accelerator.
Why it matters to me: I wanted to see how far a small, affordable computer could go in making medical imaging AI more accessible. LungCare Pi Camera is not designed to replace doctors or diagnose lung cancer. For me, it's a step toward understanding how AI can support medical image analysis, and how these technologies might eventually become available beyond well-equipped hospitals.
LungCare Pi Camera is an experimental research prototype, not a certified medical device. Its outputs are not intended for clinical diagnosis or treatment decisions.
Record breath → Mel Spectrogram / MFCC → CNN + LSTM → Risk flag
* BreathCheck AI is a research prototype and is not a certified medical device.
.jpg)
WHAT'S NEXT?
Next, I want to connect the two: let BreathCheck flag who needs a scan, and let Lung Progress Tracker follow them after it. One tool to notice, one tool to remember.