MY PROJECTS
REDDIT BDI
Designed a Web-based Application that predicted if users were depressed or not, based on their Reddit post history. Depression prediction is based on Beck's Depression Inventory (BDI) which is a form containing 21 questions. The answers to these questions determine level of depression in the user. I built a BERT model that used training data provided by the CLEF eRisk 2021 competition. The website allows users to enter in their Reddit username, the backend model then predicts the user's answers to the 21 questions and consequently assigns a level of depression.
SWAV ANALYSIS
Evaluated the performance of Facebook AI Research's self-supervised learning algorithm, SwAV (Swapping Assignments between Views), on the CheXpert dataset (a dataset of Xray images). Compared performance of a model pre-trained with SwAV versus a model pre-trained with supervised learning methods on the ImageNet dataset.
ITRUST
Lead a team of 8 to write new user stories for iTrust - a web application to manage the internal logistics of a health center. Followed Test-Driven Development (TDD) and utilized version control systems to complete this 2 month long project. Ideated and created a new user feature for the site, as well as wrote Unit and Selenium Tests for existing features.
INDOOR LOCALIZATION SYSTEM
Built a functional dead-reckoning based indoor localization system that was able to produce walking trajectories of users. Accelerometer and Gyroscope data was collected through the IMU present in the user's phone. This data was pre-processed and analyzed to then plot the user's trajectory. The three building blocks in this project were to determine step-count, walking direction and step-length.
SIMPLE DISTRIBUTED FILE SYSTEM
Designed and built a simple distributed file system (SDFS) using 10 Virtual Machines along with a new cloud-computing framework called MapleJuice. This paradigm was similar to Apache Hadoop's MapReduce. Compared performance of Hadoop MapReduce and MapleJuice on simple operations like putting and getting large files.
PACMAN
Built a version of the arcade game 'Pacman' by combining and optimizing both hardware and software on an FPGA. Languages used were SystemVerilog and C. Features included an AI-powered ghost and multi-user functionality and was interfaced with both a VGA monitor and keyboard using USB ports and the JTAG UART core.