Portfolio

Professional Experience

Business Intelligence Engineer @ TopRx, LLC (Apr 2022 - Present)

Developed and implemented a classification model using TensorFlow and SQL to generate personalized purchase lists, increasing the sales success rate by 50% for phone-based channels.
Leveraged Census geospatial data to enhance customer insights, improving product recommendations and pitch lists, resulting in a 20% increase in new item success rates and stronger customer relationships.
Created a predictive model in Scikit-learn leveraging historical trends and current customer buying habits to identify high-risk customers, enabling proactive risk mitigation strategies that reduced customer churn by 10%.
Utilized DOMO and Python to extract, transform, and visualize data from multiple sources, providing executives with comprehensive and intuitive insights that informed strategic decisions.

Research Assistant @ University of Memphis (Jun 2020 - Dec 2021)

Developed an automated markerless motion capture system to collect over 200,000 image frames of athletes’ physical motions.
Achieved 80% accuracy in identifying improper knee placement during vertical jumps using a trained KNN classifier.
Compared our markerless gait system versus QUALYSIS, identified improvements, and communicated findings to stakeholders with actionable insights.

Education

M.S., Electrical and Computer Engineering University of Memphis (December 2021)
B.S., Electrical and Computer Engineering University of Memphis (May 2020)

Certifications

Machine Learning & Data Science

SQL

Cloud Services

AWS Certified Cloud Practitioner (expires on 05/2026)

Projects

Machine Learning

SQL-based QA bot with Gemini API

Develop a chatbot capable of generating and executing SQL queries to retrieve data from a supermarket sales database and provide users with precise, context-based answers.

Main Objectives

Data Preparation: Transformed a Kaggle dataset into SQL tables, ensuring the data structure supported efficient querying and contextual analysis.
Prompt Engineering & Tool Development: Designed and tested dynamic prompts and query-generation tools to produce accurate SQL queries tailored to user inputs.
Chatbot Development: Built a question-answering system that interprets user questions, executes SQL queries, and delivers precise, contextually relevant answers.
Testing & Evaluation: Conducted extensive testing to validate the chatbot’s query accuracy, response relevance, and overall performance.

Skills: Python, Nature Language Processing, Large Language Model, Gemini API, Question Answering

Nature Language Processing: Build GPT Tokenizer from Scratch

Explore, compare, and implement the Byte-level Byte Pair Encoding (BBPE) tokenizer to demonstrate its efficiency in large language models (LLMs) by balancing vocabulary size and tokenization efficiency.

Main Objectives

Comparative Analysis: Analyzed multiple tokenization methods, including word-level, character-level, and byte-level tokenizers, highlighting the unique benefits and limitations of each approach in LLMs.
BBPE Algorithm Implementation: Demonstrated step-by-step implementation of the BPE and BBPE algorithms, from encoding raw UTF-8 bytes to iteratively merging frequently occurring pairs to build the final vocabulary.
Vocabulary and Merge Rule Creation: Built and stored the final vocabulary dictionary and merge rules, enabling efficient encoding and decoding for downstream LLM tasks.
LLM Optimization Insight: Showcased how BBPE achieves an optimal balance between vocabulary size and tokenization efficiency, addressing key challenges faced by LLMs.

Skills: Python, Nature Language Processing, Byte-level Byte Pair Encoding, Large Language Models

Computer Vision: YOLOv8 on Traffic Detection

In this project, we utilized the YOLO model for traffic detection, starting with familiarization and testing of a pre-trained model. Then, we have fine-tuned the model with our dataset. The retrained model shows a significantly improved performance. We have also analyzed different metrics and plots obtained during the training process. All of those plots confirm enhanced accuracy.

Skills: Python, Computer Vision, Model Finetuning, Metrics Evaluation

Regression Problem: House Price Prediction

Throughout the project, we have performed exploration data analysis to understand the dataset and the relationship between given features and the house price. Then, we cleaned and preprocessed the dataset. Lastly, We aimed to create a model that can predict house prices with only the provided features. As we can see from the final results, we can obtain a pretty decent performance using the random forest regression model on the unseen data.

Skills: Python, Regression Model, EDA, Data Cleaning, Feature Engineer

Data Analysis

Gait Analysis Performance Comparison

We have a running dataset that was collected using our proposed markerless gait analysis system, and another dataset was collected simultaneously by Qualisys(a marker-based gait analysis system that is the standard of the current market). We want to compare the accuracy of our system using the Qualisys system as the golden rule and find the areas of improvement.

Skills: Python, Data Analysis, Data Cleaning, Feature Engineer, Data Collection, Data Visualization

Data Visualization

Superstore Sales Overview

Create an interactive dashboard in Tableau to draw insights into the overview of superstore sales.

Skills: Tableau, Data Visualization, Reporting

Genative AI Search Trend in the US

Create an interactive dashboard in Tableau to draw insights into the search trend for different generative AI models in early 2024.

Skills: Tableau, Data Visualization, Reporting

New York Airbnb Investment by Location

By looking into the New York Airbnb records from 2008 to 2015, I built a dashboard in Tableau to enhance the decision-making process for choosing locations and properties for Airbnb investment in New York.

Skills: Tableau, Data Visualization, Reporting

Publications

Marker-less motion capture system using OpenPose

Abstract:

Motion capture systems are widely used for measuring athletic performance and as a diagnostic tool in sports medicine. Standard motion capture systems record body movement using: (1) a set of cameras to localize body segments; or (2) specialized suits in which inertial measurement units are directly attached to body segments. Major drawbacks of these systems are limited portability, affordability, and accessibility. This contribution presents a markerless motion capture system using a commercially available sports camera and the OpenPose human pose estimation algorithm. We have validated the proposed markerless system by analyzing the human biometrics during running and jumping movements. The findings of this study demonstrate that pairing a low-cost sports camera with artificial intelligence allows for highquality analysis of human movement.