Data Scientist, Security Data Science
Machine learning for fraud, threat and privacy problems, and the data engineering that makes it trustworthy.
Portfolio · CV · LinkedIn · Email
- Ph.D. student in Computational Sciences and Engineering at Harrisburg University of Science and Technology (since October 2026).
- M.S. in Cybersecurity from Yeshiva University, Katz School of Science and Health (CGPA 3.966).
- AI Lead Researcher at UsHealthCloud, working on in-enclave LLM fine-tuning for oncology research on Trusted Execution Environments.
- Author of five papers on federated learning, deepfake detection, fraud and intrusion detection. Nominated Cybersecurity Innovator of the Year (2026 Cybersecurity Excellence Awards).
The first two run in the browser, so you can try them. Each repository has tested code and a README with the full results.
| Project | What it shows | Headline result |
|---|---|---|
| Phishing URL detector · live demo | Honest evaluation, LightGBM in the browser, explainable scores | 84% of live OpenPhish links caught at 2.1% false alarms on unseen sites; exposes a dataset shortcut behind 99.9% scores |
| Small-count privacy audit · live demo | Privacy attacks, linear programming, differential privacy | Recovered 216 "hidden" census counts from published totals; DP at ε = 1 costs 0.77 people per cell |
| Mobile-money fraud detection | Leakage audit, LightGBM, real-time FastAPI scoring | 99.7% of fraud caught at 64% precision; explains why PaySim models score near 100% |
| Live Wikipedia lakehouse | Streaming, Apache Iceberg, change data capture for RAG | Edits searchable in 17 s; 96.5% fewer embeddings than re-indexing; 0 duplicates after crashes |
| LLM pretraining data refinery | Deduplication, decontamination, PII redaction | 339k raw web pages to 62.1M clean training tokens in about 6 minutes on a laptop |
| FinanceBench RAG retrieval | Retrieval evaluation, learning to rank | Evidence recall@5 from 7% to 58% over 84 SEC filings |
| Demand forecasting and inventory | Probabilistic forecasting, decisions under uncertainty | 19% better WRMSSE than AutoETS; about $113k a year more profit |
| Uplift targeting | A/B testing, causal ML, off-policy evaluation | Showed a targeting model should not ship yet; found the $0.27-per-contact break-even |
| Olist analytics engineering | dbt, SQL, data quality, causal analysis | A late delivery adds 46 points of 1-star risk; 63 sellers cause 13% of late orders |
- A Federated Learning Approach to Secure AI-Based Patient Outcome Prediction Across Hospitals (accepted 2025)
- Mitigating Deepfake-based Impersonation and Synthetic Data Risks in Remote Healthcare Systems (published 2025)
- Machine learning and Blockchain approaches for enhancing fraud prevention in financial transactions (published 2025)
- Mitigating Ransomware in the Energy and Healthcare Sectors through Layered Defense Strategies (accepted 2025)
- The Role of AI in Strengthening Privacy and Security in the Era of Cyber Crime (accepted 2025)
| Languages | Python, SQL, Rust, C# |
| Machine learning | LightGBM, scikit-learn, SHAP, EconML, statsmodels, causal inference, forecasting |
| AI and LLMs | LLM fine-tuning, RAG, embeddings, vector search, LLM training-data pipelines |
| Security and privacy | Federated learning, differential privacy, Trusted Execution Environments, threat detection, Zero Trust, NIST 800-53/171 |
| Data engineering | pandas, Polars, DuckDB, dbt, Dagster, Apache Iceberg, streaming and CDC |
| MLOps and cloud | FastAPI, Docker, Kubernetes, GitHub Actions, AWS, Google Cloud, NVIDIA GPUs |
Certificate of Cloud Security Knowledge v5 · NVIDIA Certified Associate · Certificate of Competence in Zero Trust · Red Hat System Administration · ISC2 Cyber Pathway · AI in Healthcare Specialization · ISACA member
I am open to data science roles in security, fraud, privacy and AI safety, in New York or remote. The quickest way to reach me is KMuhwati@my.harrisburgu.edu.