Personal website with a few of my projects over the years
Please have a look at some of my projects below. My GitHub profile and resume are linked on the left.
This project focused on exploring the way in which humans express humor. Through an analysis of text, audio, and video, we employed a modified Transformer architecture to encode the features. We were able to define a binary humor prediction task on the test data and achieved state-of-the-art results for our task and dataset. We showed that gesture, as extracted from video, is a salient feature informing the expression of humor in an individual. (final paper)
This project focused on reproducing and expanding on recent work in orthogonal constraints in structural probes (original paper). The primary research focuses on using structural probes for BERT embedding interpretability. We were able to reproduce the results, particularly showing that orthogonal constraints are less prone to memorization of training data when compared to unconstrained probes. We also extended this research with ablations studies exploring different distance metrics, loss functions, and evaluation metrics. (final paper)
This project focused on developing real-time methods for detecting methods for empathetic response. We focus on both classification of disclosure and generation of the responses. The project uses several natural language techniques (such as BERT and RoBERTA embeddings) to fine tune existing models and train models from scratch. This is an ongoing project so updates will be shared with progress.
This project focused on developing a system to be used for robot mindfulness interactions. We designed a system to measure heart rate (HR), inter beat interval (IBI), and heart rate variability (HRV) through live remote video (Zoom). We used photoplethysmography to capture these signals and applied a variety of denoising techniques to purify our desired measures. In addition, I designed the data collection pipeline which included a decoupled multi-process architecture that allows for data collection, data processing, and data viewing in realtime. Source code can be found here.
This project focused on developing a SQL based robot logging system that allows for quick topic based information storage to be used in large-scale robot deployments. In addition, the topic-based logging enforced datatype constraints on fields which allows for robust data storage and a graceful handling of errors. The tool is publicly through MIT open license and is available on PPI. Source code can be found here.
This project focused on creating a conversational agent that interfaces directly with the Amazon Lex chatbot. I developed a python wrapper around the Amazon API that allows for quick deployment of conversational agents, addition of new slots and intents, and importing existing projects. Source code can be found here.
This project focused on developing a hands-free projector interaction system. I developed a system that utilizes structured lighting cameras to estimate skeletal pose, allowing for programs to use hand movements as inputs to a projected surface. This allows for interactions that do not require any special wearables. Source code can be found here.
This project focused on developing Open Home, a privacy-first smart speaker with an interactive virtual assistant that can play music, control lights, and tell time. I served as project lead for a team of six students during this semester long project which included interfacing regularly with our stakeholder to ensure the MVP would be completed in a timely manner that met expectations. We designed data security protocols including local storage solutions, shielded internet exposure, and decoupled hardware/software control. In addition, we streamlined feature creation with modular widget structure, allowing users to quickly deploy their own functionality with a few additional lines of code. Source code can be found here.
This project focused on creating a Markovian haiku generator that scrapped news websites for content. It was developed as an experimental poetry project for a writing course. I also included a graphical user interface that allows users to enter search keywords to be used as the content source. Source code can be found here.