Click on a project card to redirect to the project URL!

BEReddiT

Analyzed controversy in Subreddits and designed a data visualization dashboard to display results
- Modified and maintained Subreddit data scraper with Asyncpraw
- Designed pipeline to preprocess subreddit data and feed into BERT model to get topics and sentiment analysis on subreddits
- Wrote a module to get comment relevancy to their post
- Designed “controversy” metric from topics, sentiment, and relevancy to rate “controversy” of a subreddit
- Developed data visualization dashboard to show results of topics, sentiment, relevancy, and controversy using Plotly Dash

Evaluating the Robustness of Collaborative Filtering Recommender Systems against Attacks

Our goal is to analyze the five shilling attack methods: Random Attack, Average Attack, Bandwagon Attack, Segmented Attack, Sampling Attack on different recommender systems. Recommender models chosen for evaluation are: User-based KNN model, Item-based KNN model, Latent Factor model, Neural Matrix Factorization model
- Implemented functions for Random Attack, Sampling Attack, and Segmented attack
- Implemented Neural Matrix Factorization model and trained on different attacked data sets
- Discovered KNN and Latent Facor recommender systems are more vunerable; Neural Matrix Factorization is the most robust against attacks

Benchmarking Azure and AWS spark sessions by K-Means

My goal for this project was to benchmark Azure and AWS spark sessions.
- Implemented a K-Means function for spark RDDs
- Ran the running time of K-Means on Azure and AWS with varying sizes of data and number of workers
- Discovered that Azure spark has a slightly faster running time than AWS
- Discovered size of data impacts running time linearly and number of workers impacted running time exponentially

Trainer AI

A project that can train people to learn correct postures for sports. This project uses computer vision to learn the correct postures from a professional sports player. A student can try to mimic the professional sports player's movement, and the program will give a score for how well they performed in comparison.
- Built front-end flask app to upload trainer video and watch students live
- Extracted human body positions using mediapipe
- Implemented and refactored functions to recognize poses from trainers and students

Recipeat

Recipe recommender for Meal Planning based on nutrition and available ingredients.
- Developed end-to-end Multi-Touch Attribution Model to attribute credit to marketing campaigns for a customer purchase
- Designed modules to query data, clean the data, model the data, and generate a visualization for the user
- Saved marketing spending on underperforming marketing campaigns while maintaining optimal customer purchase
- Designed and developed a trend anomaly diagnostics system which assisted in understanding how and why an anomaly occurred; enabling users to interpret and take action when facing a detected trend anomaly alert

COVID-19 and Government Policies

Find which government policies were the most effective at slowing down COVID-19 cases
- Searched and scraped data online regarding COVID-19 cases and government policies implemented in the United States
- Cleaned the data into a usable format for machine learning algorithms by using Pandas package from Python
- Analyzed government policies from generated auto-regression time series and SAS JMP graphic visualizations and removed nonimpactful government policies from the regression
- Discovered the most impactful government policies were mask mandates and gathering restrictions of any size

Android Chess App

Created an Android App to play chess with another human player or a bot
- Designed graphical user interface using Android Studio
- Developed code to play Chess through console output
- Integrated the Android graphical user interface to be compatible with Chess code
- Enhanced App by adding extra functionality such as playing against AI, saving games, and recording previously played games