Case studies

Three production systems where the interesting work was the infrastructure, not the prompt. Each entry below is a deep dive on architecture, decisions, and the things that broke.

All projects

Production · Regulated nonprofit

Multi-model extraction pipeline

Configurable multi-stage agent pipeline with field-level consensus voting. 30× throughput, eliminated a full-time contract position, every decision auditable end-to-end.

Production · ~600 employees · 20 countries

Enterprise AI governance platform

Use-case registry with four-tier RBAC, council review workflow, native Teams + SharePoint embedding. Built so adoption tracking and compliance review run against the same surface.

Active development · Open architecture

Spire

AI-native infrastructure control plane in Rust. Policy-gated approval flows, per-object audit log, MCP-based tool access — single-binary deploy, no master keys handed to agents.