Production ML, NLP and LLM-orchestrated systems.
A data scientist working where models, pipelines and the people who rely on them meet.
For the past four years I've built data systems that have to hold up in production, not just in a notebook. That work spans healthcare, job-market data and university operations.
At the University of Maryland Counseling Center I built a no-show prediction model on HIPAA-regulated clinical records, wrote the DuckDB pipelines underneath it, and built the dashboards the counseling staff open every morning. At Canaria I ran anomaly detection over 15M+ job-market records and a salary model across 40,000+ US zip codes.
The tools change. What hasn't is the approach: understand the problem properly, be suspicious of your own data, and make the result something a person can act on.
A model nobody uses isn't finished.
An agentic trip planner that orchestrates seven MCP servers through Claude, pulling live flights, hotels and events into a single request. I load-tested EC2 against ECS Fargate at 50 to 100 concurrent users; Fargate matched EC2's 460 RPS on half the compute.
A YOLO vision pipeline that spots wildlife on farmland in real time and fires a non-lethal deterrent matched to the animal, so nobody has to watch a monitor. Hardware prototype built and tested alongside the model.
End-to-end pipeline: live BTC collection, ARIMA and XGBoost models at around 85% accuracy, automated ETL, and Qlik Sense dashboards giving real-time visibility into volatility. Cut the manual reporting effort by 40%.
A full analytics pipeline over a multifamily portfolio: SQLite to engineered features to five business reports, a seven-sheet Excel workbook and Power BI. Surfaced $2.14M of annual loss-to-lease and a 26-point occupancy gap. The data is synthetic, generated with Faker. This was built to demonstrate the pipeline, not to report on a real client.
Fig. 003 · Four projects, ordered by how much I learned rather than by date.
Open to full-time data science, ML engineering, and analytics roles. The inbox is always on.
An AI answering as me, from my real resume and projects.
