This project combines a mobile prototype with an ML screening pipeline for respiratory audio. The model compares cough and breath samples using pretrained embeddings (YAMNet and wav2vec 2.0) and highlights confidence bands for rapid screening.
Nikhil collected consented audio samples, documented clinical consent procedures, and tested the prototype in a community clinic setting. The resume photos from those testing sessions are included in the gallery. The result is a practical, low-friction tool that connects clinical workflows with modern audio ML.
Gallery
