I'm an economist turned data scientist who spent 12 years building a million-dollar coffee business while earning my MA in Economics. Now I build predictive models and applied AI systems, with a particular interest in causal inference and the kind of machine learning that earns trust in high-stakes environments.

Economics taught me that stories are rarely simple and every decision has hidden costs. Machine learning lets me quantify those trade-offs with precision. At Veterans United Home Loans, I developed delinquency prediction models for multi-billion dollar mortgage portfolios that replaced grid-based systems with continuous modeling. The old approach used discrete FICO score bins where a loan with a 759 score was valued identically to one with a 740 score, but a 760 score received a substantial jump in valuation despite minimal marginal risk difference. My continuous models revealed the hidden cost: suboptimal keep-versus-sell decisions that left money on the table when meeting portfolio retention targets.

My current work is at the intersection of machine learning and state government systems. As a Business Analyst with Missouri's Office of Administration supporting DESE applications, I've traced and refactored legacy code that routed 300,000+ record-level database calls through chains of SSIS packages and intermediary applications — and replaced them with array-level SQL queries. On my own time, I built and deployed a production RAG system querying 22.6 million words of Missouri education policy, accessible at rag.attwoodanalytics.com. The system uses GPU-accelerated embeddings, a domain-specific glossary with query-time acronym expansion, hybrid vector and keyword search, and citation enforcement. Building it taught me that corpus curation matters more than raw volume: systematic quality filtering reduced 38,199 raw chunks to 2,599 high-signal chunks and directly improved response quality.

I'm drawn to problems where predictive models and intelligent agents can do real work in production — not demo environments. I want to know not only what works, but why it works. That perspective becomes especially valuable in applied AI development, where trust in the system has to be earned through interpretability, reliability, and honest evaluation.