My work focuses on document processing, workflow automation, and data systems where accuracy, traceability, and human review matter.
At Foley, I build tools for teams doing high-stakes, repetitive work. These projects automate the parts software can handle and make it clear when a person needs to step in.
Operations teams were splitting 50- to 75-page driver packets by hand, then matching each document to the right driver. I built a system that uses simple layout rules when the split is obvious, brings in vision only when it is not, and sends uncertain results to a reviewer.
Customer churn affects revenue, so Customer Success needs a clear way to decide which accounts may need attention first. I built a transparent scorecard from 43 months of history, using only the information available at the time of each score. In one segment, the highest-risk 20% captured about 55% of later churn events. The testing also showed that two signals people expected to matter did not hold up.
Screening teams had to search court by court, sort out people with similar names, review PDFs, and re-enter what they found. I built one workspace that searches dozens of federal district courts. Rules decide which records become candidates, every finding links to its source page, and a person—not the model—makes the compliance decision. The system still works without the model.
Machine learning and research projects, with the code on GitHub.
A PyTorch chess bot trained through self-play, Monte Carlo tree search, and neural network evaluation.
View repositoryML researchFeature selection and classification on miRNA expression data from prostate cancer samples.
View repositoryComputer visionReads a hand-drawn chessboard and converts the position into FEN notation.
View repositoryAI productSearches multiple job boards, extracts structured listing data, and removes duplicates.
View repositoryI build AI tools for document processing, workflow automation, data migration, and other compliance-sensitive operations.
Python · TypeScript · PostgreSQL · LLMs · OCRI developed an IP-to-FQDN transformation engine and a DNS analysis pipeline for research on network topology and naming patterns.
Python · pattern recognition · data pipelinesI built feature-selection and classification workflows for prostate cancer miRNA data using XGBoost, support vector machines, and transformer models.
PyTorch · XGBoost · scikit-learn · researchI designed and shipped a responsive dashboard for customer support reporting, monitoring, and data visualization.
React · TypeScript · product engineeringA model is only useful if people can understand the result and know when to review it.
I work mainly in Python and TypeScript, with SQL and PostgreSQL for data systems. My recent projects have included OCR, structured extraction, LLM integrations, retrieval, classification, and workflow automation. When a system can be wrong, I add validation, preserve an audit trail, and make it easy to hand uncertain cases to a person.