Convolutional neural network that classifies thin blood smear slides as parasitized or uninfected. The model is a small VGG-style CNN trained on the NIH malaria cell dataset, exposed through a Flask web app that accepts an image upload and returns the prediction with the confidence score.
Collection of Jupyter notebooks covering classical machine learning and deep learning building blocks. Each notebook walks through one technique end to end, from data preparation to evaluation, using scikit-learn, TensorFlow, and PyTorch on small public datasets that fit in memory.
Direction-finder for visually impaired pedestrians. The app pulls live camera frames, runs a MobileNet object detector on device, and converts detected obstacles into spoken cues through the platform speech synthesiser. Built as a Flutter front end against a TensorFlow Lite model.
Mobile app that helps visually impaired users perceive their surroundings. The camera frame is run through a MobileNet object detector and the results are read aloud through the system text-to-speech engine. Won the MLH Fellowship Sprint 0 hackathon during the Fall 2020 cohort.