Data science that lives inside real workflows.

CUNY Institute for State & Local Governance

Built applied ML and data infrastructure for legal-data standardization and nationwide policy research.

• Architected a hybrid AI reasoning platform for legal charge standardization across 16 U.S. jurisdictions, combining dense retrieval, Sentence-BERT embeddings, knowledge graphs, DSPy-optimized prompting, and multi-agent LLM workflows.
• Developed calibrated ML and deep-learning ensemble decision systems with LLM adjudication, reaching 95%+ automated classification agreement.
• Built reproducible 2014–2024 ACS pipelines for all U.S. counties using Python, DuckDB, Polars, dbt, Delta Lake, Arrow, Dagster, and Great Expectations.

NYC Office of the Mayor

Applied machine learning, retrieval, forecasting, and analytics to procurement and M/WBE compliance workflows.

• Improved compliance risk detection by 32% and reduced manual audit effort by 45% using classification and anomaly detection.
• Built a RAG platform over policy, contracts, compliance records, and agency guidance that reduced policy lookup time by 78%.
• Improved procurement utilization forecasting accuracy by 24% across 30+ city agencies.

Baruch College, CUNY

Supported 80+ students in Service Operations Management and Lean Six Sigma, covering forecasting, regression, SPC, process capability, DMAIC, and applied statistical reasoning.

Gentrainer, Gentrainer, Meyers Workforce Solutions, LLC.

Built quality-scoring and governance workflows for AI training data and agent outputs.

• Reduced noisy AI training labels by 31% and improved prediction precision by 23% with Python, SQL, XGBoost, and deep-learning evaluation.
• Reduced manual QA effort by 42% and validation time by 45% through 103 engineered features and MLflow-based governance, explainability, drift detection, and benchmarking.

Think & Learn (BYJU'S)

Built causal ML and decision systems for personalization, counseling interventions, and acquisition efficiency.

• Increased incremental student conversions by 17% using Double Machine Learning with EconML, DoWhy, XGBoost, Python, and SQL.
• Reduced customer acquisition costs by 39% through uplift modeling, policy simulation, counterfactual analysis, and decision intelligence.