Projects
A selection of applied projects focused on decision support, interpretability, and real-world constraints.
MoneyTrail — Fraud Analysis & Risk Investigation System
MoneyTrail is a fraud analysis system designed around how analysts actually investigate risk, not just how anomalies are detected.
Account overview with risk score, flagged behaviors, and metadata surfaced in one view.
Ranked accounts table prioritizing high-risk entities for rapid triage.
Detailed account view highlighting behavioral patterns and triggered flags.
Simulated deposits, withdrawals, and transfers to observe risk propagation.
Suspicious activity feed ordered by severity and time.
Entity network graph revealing hidden relationships across accounts and merchants.
Rather than surfacing every suspicious event, I focused on ranking signals, showing relationships, and helping users quickly understand why something looks risky and what to examine next.
Tools: Python, SQL, network analysis, interactive dashboards
View on DevpostElectra - Voter Simulation
Electra is a simulation platform that explores how assumptions and narratives influence voter behavior under different scenarios.
Instead of producing a single prediction, the system emphasizes explainability, showing how changes in inputs lead to different outcomes and making uncertainty explicit.
Tools: LLM agents, scenario simulation, data visualization
View on DevpostFannie Mae
Data Engineering & Applied Machine Learning
Working on production systems changed how I think about machine learning. Models were only one part of a larger workflow that included data quality, interpretability, and downstream decision-making.
That experience reinforced the importance of building systems that hold up outside of notebooks where reliability and trust matter as much as performance. Because this work involved internal tools and sensitive data, I’m not able to share screenshots or code publicly.
Focus: Production pipelines, model evaluation, analyst-facing systems
Explore more: