A breath test for your gut

A handheld breath analyser for gastrointestinal screening, aimed at conditions like SIBO and IBS. It was my final-year capstone at Thapar Institute, built by a team of three: Daisy, Akashdeep Singh and me, mentored by Dr. Maninder Kaur.

• Three gas sensors on an ESP32: MQ-8 for hydrogen, MQ-2 for methane and MQ-137 for ammonia, in a 3D-printed enclosure that directs breath across them.
• The ESP32 streams readings over Bluetooth Low Energy to an Android app written in Kotlin and Jetpack Compose, with Room over SQLite on the phone.
• A stateless Flask API in Python scores each reading in under 50 ms.
• The model is a Random Forest, 100 trees in scikit-learn, trained on 1,200 synthetic samples, 200 per diagnostic category.

Clinical threshold rules run deterministically alongside the model, so a prediction on its own is never the output.

The model scored 85.8% on held-out test data. That data was synthetic. There was no patient data and no clinical validation, the report says so more than once, and so does this page.