Data Scientist at T-Mobile | Fraud & Credit Risk | GenAI & LLM Systems
6+ years building end-to-end ML pipelines that catch fraud, optimize credit risk, and save real money — $1M+ in monthly loss prevention from models I designed, deployed, and monitor in production. I work across the full stack: from feature engineering on multi-terabyte data (PySpark, Snowflake, BigQuery) to deploying LightGBM models on Databricks, to building LLM-powered tools with LangGraph and RAG.
Previously: Data Scientist at DEPT (media mix modeling, $30M+ ad spend reallocation), Data Analyst at Tekion (recommendation engines), Graduate Research Assistant at Georgia State University (NLP classifiers).
- Production ML — fraud detection models scoring 500K+ transactions/month, cascaded scoring pipelines, champion/challenger A/B testing frameworks
- LLM Applications — enterprise RAG agents, multi-agent systems with LangGraph, hybrid LLM + deterministic pipelines with structured outputs
- Causal Inference — GeoLift incrementality testing, treatment-assignment frameworks, causal guardrails for experimentation
| Project | What It Does | Stack |
|---|---|---|
| Hirelens | AI resume analyzer — scores resume-to-JD fit, diagnoses rejection stage, generates fixes. Hybrid LLM + deterministic pipeline with eval harness. | FastAPI, Groq (Llama 3.1), Pydantic, Chart.js, Docker |
| Collections Intelligence Agent | Multi-agent system that translates plain-English strategy questions into SQL, evaluates performance, delivers executive reports. | LangGraph, Streamlit, SQLite, Docker, GitHub Actions CI/CD |
Languages & Data: Python, PySpark, SQL, R, Pandas, NumPy, Scikit-learn
ML & Modeling: LightGBM, XGBoost, PyTorch, TensorFlow, SHAP, Causal Inference, A/B Testing
GenAI & LLMs: LangGraph, LangChain, RAG, Prompt Engineering, Groq API, Anthropic Claude API, Hugging Face
Cloud & MLOps: Databricks, AWS, GCP, Azure, Docker, MLflow, GitHub Actions CI/CD
Data Platforms: Snowflake, BigQuery, PostgreSQL, MongoDB, Delta Lake
Visualization: Tableau, Power BI, Plotly, Streamlit
