Most of my work starts with a workflow nobody has specified cleanly.

I figure out how the work actually happens, identify the decisions and failure modes that matter, and build whatever the result requires. That work has included industrial catalog resolution, revenue intelligence, customer operations, developer tools, and agent infrastructure.

I work at whatever layer is constraining the result. Sometimes that is the interface or data model; sometimes it is a service, model boundary, integration, deployment, or recovery path. The layer matters less than whether the whole system works.

I use models where evidence is ambiguous. Identity, permissions, budgets, state changes, and external effects stay in explicit code. I put acceptance criteria beside the work that must satisfy them, preserve the evidence behind consequential results, and turn failures into regression tests.

I build and deploy agents across CRM/ERP systems that help manufacturers automate and scale time-consuming, error-prone workflows in sales ops, revenue ops, customer service, and manufacturing ops (quoting, order management, demand forecasting, inventory management, purchasing, job routing, and more).