Recipe-Restaurant Web App |   September 2020 – December 2020
Design and build a website with a database cross-referencing all dishes across Yelp restaurants and publicly available recipes and/or ingredient lists. [MySQL, REACT, Python]

Time series model |   June 2020 – August 2020
Created seasonal and trend-sensitive time series model detecting and analyzing Mercedes-Benz server outages and data logging failures using seasonal decomposition, ARIMA and SARIMAX, while investigating other approaches including LSTMs and Prophet. [Python, PySpark]

Standardized Testing AI |   March 2020 – May 2020
Explored the effects of transformer model distillation on the question answering task. In particular, we fine-tune BERT, RoBERTa, distilBERT, and distilRoBERTa on the Stanford Question Answering Dataset (SQuAD) and compare model performance, model training time, and model inference time. [Python, Tensorflow]

AI Chatbot |   November 2018 – March 2019
Fine-tuned and published the trained BERT model as a service to build a Question Answering system - revolutionizing the AI chatbot - that is able to act as a real-life operator for LogMeIn’s Bold360 AI’s clients. (won 3 prizes at regional Hackathon) [Python, Tensorflow]

Youtube Spam Detector |   November 2017
Classified comments gathered from Youtube as ham or spam with Naïve Bayes and SVM models by using TF-IDF transformation. [Python, scikit-learn]

County Election Predictor |   September 2017
Built Neural Network model in Keras on top of TensorFlow to predict if Clinton or Trump wins in the 2016 election broken down by counties. [Python, Keras, TensorFlow]

UPenn logo detection |   May 2017
A computer vision project where I fine-tuned an image classification model trained on the ImageNet Large Visual Recognition Challenge dataset. My goal was to identify images which included the UPenn logo from those of not containing the logo by populating training data (collected from the internet as well as photographs taken by hand) through Python scripts. [Python, TensorFlow]

Raspberry Pi Security System |   May 2016
Utilized motion detector to initiate the security system to take a picture and trigger email response system. [Python, Shell scripting]