addPart-Time Assistant Professor of Data Science & Analytics at Kennesaw State University. I study when AI-generated data can be trusted, with a focus on LLM agents, synthetic respondents, demographic fidelity, and data integrity.
AI systems are increasingly used to generate synthetic data, automate data collection, and stand in for human respondents. My research asks when those outputs are valid, representative, and trustworthy. I build LLM agents for structured data collection, evaluate memory and statefulness, audit synthetic populations against real-world benchmarks, and study how AI-generated submissions can be identified when automation threatens data integrity.
My work sits at the intersection of trustworthy AI, synthetic data generation, survey methodology, and algorithmic bias. Alongside my academic research, I have partnered with Delta Air Lines, Blue Cross Blue Shield of Tennessee, Atlanticus, and Southern Company on applied machine learning and analytics problems in transportation, healthcare, finance, and energy.
I developed a locally deployed LLM agent that combines browser automation, DOM parsing, COSTAR-A prompting, layered response matching, and validation-error correction to complete multi-format Qualtrics surveys without human intervention.
I compared no memory, rolling summarization, structured JSON injection, and retrieval-augmented generation across surveys of 10 to 100 questions, measuring internal consistency, token cost, and completion time.
I audited 10,138 synthetic personas from ten contemporary models across the Claude, GPT-5, GPT-OSS, Llama, and Qwen families against U.S. Census benchmarks under batch and incremental generation.
I currently serve as a Part-Time Assistant Professor in Kennesaw State University's School of Data Science and Analytics, teaching statistics in R and statistical data mining in Python. My broader teaching experience spans mathematics, statistics, and data science at the undergraduate and graduate levels in the United States and Nigeria. Roles marked IoR were taught as instructor of record.
| Course | Level | Role & Term |
|---|---|---|
| Statistical Data Mining DATA 4310 | Bachelor's | IoR · Fall 2026 |
| Statistical Methods I STAT 3120 | Bachelor's | IoR · Fall 2026 |
| Introduction to Data Science DATA 1501 | Bachelor's | IoR Spring 2026 · TA Fall 2023 |
| Python for Data Science DS 7140 | Master's | TA · Spring 2025 |
| Statistical Computing STAT 7020 | Master's | TA · Fall 2023, Spring 2024 |
| Data Mining in Business Analytics STAT 4440 | Bachelor's | TA · Spring 2022 |
| Data Mining STAT 6440 | Master's | TA · Spring 2022 |
| Regression Analysis STAT 5020 | Bachelor's, Master's | TA · Fall 2021 |
| Linear and Integer Programming OR 6610 | Master's | TA · Fall 2021 |
| Elementary Differential Equations I & II MTH 203 / 204 | Bachelor's | IoR · 2019-2021 |
| Elementary Mathematics I & II MTH 101 / 102 | Bachelor's | IoR · 2019-2021 |
addSelected as one of two winning posters for “Comparing Memory Architectures for LLM-Based Synthetic Survey-Response Generation.”
addRecognized as the top doctoral student in KSU's PhD in Data Science & Analytics program.
addGraduate College, Kennesaw State University. Awarded $1,000 for communicating doctoral research to a non-specialist audience in three minutes.
addRecognized for research on NLP-based early warning detection of geopolitical oil supply shocks.
Proposal examining the availability of benchmark data for evaluating the validity of AI-generated synthetic survey respondents in low- and middle-income countries.
Proposal for research strengthening trust in internet systems through resilience, reliability, accountability, and security at the application layer.
Proposal supporting work that builds capacity to address risks and research challenges associated with transformative artificial intelligence.
add
add
add
addAnalyzed 32M+ calls, 13M+ messages, and 60M+ intent records to identify drivers of repeat customer contacts. A CatBoost model reached 77% accuracy and supported recommendations for a modeled 13% reduction in repeat contacts.
addForecast early customer churn across 64 engineered features. A random forest reached 96% accuracy and 96% F1.
1st place · CDSA Industry CompetitionAnalyzed 540K+ member records and 487K+ medical claims to evaluate continuous glucose monitoring, including case-control matching that produced more than 7,500 matched pairs.
Modeled credit risk on 988K+ customer records, applying SMOTE for class imbalance and reducing 458 variables to 116. A random forest reached 88% accuracy and 81% precision.
3rd place · CDSA Industry CompetitionFive classifiers across 5,574 text messages. Random forest reached 98.1% accuracy.
GitHubTransfer learning with MobileNetV2 and Bayesian hyperparameter tuning on 1,200 images, reaching 90% test accuracy.
GitHubFive classifiers on European cardholder data. A neural network reached 96% accuracy.
GitHubA PSOAANN one-class model built on Spark, running over 10M+ transaction records.
GitHubA coupon-usage recommender built from ensemble and boosting classifiers with competitive accuracy.
GitHubLongitudinal multilevel models across 29 major airports, with airport-level and seasonal patterns examined over time.
GitHub