CMS Medicare Billing Analysis
Scope: Exploratory analysis of Medicare Part B billing data
Focus: Service volume, cost distribution, provider segmentation
Highlights:
- Identified skewed cost patterns and multi-modal payment behavior
- Engineered features for outlier detection and clustering
- Built dashboards to visualize billing trends across states and procedures
Tools: Python, Pandas, Seaborn, Plotly, Jupyter
Screenshots:Coming soon
Cybersecurity Malicious Session Tracker
Scope: Exploratory analysis of Locked Shields exercise data
Focus: Identifying malicious network sessions hidden among benign traffic
Highlights:
- Detected extreme class imbalance (~2% malicious sessions) requiring careful preprocessing
- Built internal dashboard to track, filter, and visualize session metadata and threat indicators
- Enabled rapid triage and pattern recognition across high-volume network logs
Tools: Python, Pandas, PHP, JavaScript, MySQL, scikit-learn
Screenshots: Coming soon
Northwestern Terror Early Warning System
Scope: Behavioral analysis of terrorist organizations
Focus: Predicting attacks, identifying high-risk conditions, and informing policy
Highlights:
- Built an internal web application to gather data and semi-automate behavioral analysis across six terrorist groups
- Developed predictive models to forecast target types and attack likelihood over 1–7 month windows
- Released monthly public reports covering Boko Haram, Lashkar-e-Taiba, Abu Sayyaf, Al Shabaab, JNIM, and Indian Mujahideen
- Built a Web Application used to gather data and semi automatically run the analysis
- Published Machine Learning Techniques to Predict Terrorist Attacks (May 2025)
Tools: PHP, JavaScript, MySQL, Python, scikit-learn, Java
Reports: Subscribe here with an official email address
Screenshots: Coming soon
Global Online Deepfake Detection System (GODDS)
Scope: Detection of synthetic media across audio, video, and image formats
Focus: Assessing the likelihood that a given clip is a deepfake
Highlights:
- Developed a classification pipeline to flag suspicious media using supervised learning
- Built a dashboard, currently only available to journalists, to upload and evaluate content
- Currently available to verified journalists via institutional registration
Tools: PHP, JavaScript, MySQL, Python, scikit-learn
Access: Sign up here with a work email address