This page holds my portfolio of evidence, towards Ph.D candidacy in Computer Science. My research focuses on image classification, in particular, active learning in the presence of label noise. My ultimate goal is to improve state of the art methods on current benchmarks in this domain, then go on to implement these methods in products with large scale impact.

Mblessse is a hub that allows college students to localize ideas, event information and resources that are either scattered all over the internet, or clustered within campus cliques and hence had to find per college.

Being from Lesotho and studying in America, I found it rather annoying that I couldn't get friends and family back home airtime top-up. So I decided to quickly build a web app to facilitate this and uses Paypal for the payments service.

So as a pet project I decided to build an image based intruder alert system for my dorm room (LOL). Things it can detect are door, opening, chair and person. Also sends you an email with the intruder's picture as they open the door.

Bowride is a website that will allow users to organize carpools with other members of the UH community. Users can sign up to be a driver or rider, and drivers can search for riders that are most convenient to/from campus. This was a group effort for the ICS314 (Software engineering) final project.

In this project, A framework for approximating known hierachical clustering algorithms using machine learning for topic modelling is presented.

The final model implemented is Siamese LSTM to classify pairs of sentences as either the same question or different. the model is then used in a Unity VR app to help students improve on their presentation skills. This was for the ICS685 project.

A bidirectional encoder-decoder LSTM neural network is trained for text summarization on the cnn/dailymail dataset.