I build identity and authorization infrastructure. For the last three years I've been shipping AI systems into production inside a global enterprise. My working thesis: as agents take over more of the work, identity, authorization, and audit become the layer everything else depends on.
Twenty-five years building enterprise software, including identity infrastructure (OAuth, OIDC, SAML, Ping Identity), production AI systems, and B2B commerce. MPS, NYU Interactive Telecommunications Program.
Currently working on: delegated authorization for AI agents, OAuth and token exchange, auditability, and the infrastructure that makes agent deployment survivable.
Architected and shipped the internal AI platform for a global publishing house: FastAPI, React, Azure OpenAI, RAG over internal content, JWT-based access control, and per-user usage logging. In production across editorial and business workflows.
An automated code review and repair engine that scans applications, classifies findings by repairability, and applies policy-governed fixes autonomously. The core design problem is authorization: which agent actions are safe to auto-approve, which need a human, and how to make that decision auditable. Includes WCAG 2.1 AA accessibility scanning via Playwright, extends to security and code quality through pluggable reviewers, outputs SARIF for CI, and runs as a Claude Code plugin so findings land in the developer's existing workflow.
Migrated two legacy authentication systems into a unified Ping Identity / DaVinci architecture serving four enterprise applications, including the delegation and flow design for B2B commerce login and user management.
Senior engineer on the rebuild of a nine-figure B2B ordering platform: headless commerce, checkout and authorization flows, and the integration seam between commerce and identity.
A research and writing vault framework: articles, sources, talk pages, and a librarian layer for compaction and navigation. Built for thinking in public with provenance intact.
Agent-first authoring surface that treats a web app as a graph of components, props, and effects.
A multi-agent simulation framework implementing active inference (FastAPI, Mesa, PyMDP, Next.js). Agents with resource dynamics operate a simulated economy; the substrate for the stability research below.
This grows out of the production work above: the control structures that make long-lived agent systems safe, legible, and governable.
Alongside my engineering work, I pursue these questions through independent research. Current work explores mechanisms for maintaining predictable behavior over extended interactions, and models of human–agent interaction designed for deliberate use rather than engagement optimization. One project is patent pending; the others are evolving through prototypes and technical writing rather than public repositories. Together they inform how I build production systems: observable, auditable, and designed for trust.