The problem
While doing title production system implementations, we kept seeing the same pattern: title agencies with solid revenue and almost no visibility into their own operations. The data existed, inside ResWare, RamQuest, and other systems, but turning it into reports people trusted was a job nobody on staff had time for.
Ops Insights was built to be a plug-and-play business intelligence department: a library of reports for each production system, templated roles, alerts when key numbers move, and custom report writing when the library doesn't cover it.
What makes it hard
Two companies can both run the same production software and have completely different data underneath. Configuration, naming conventions, and workflows vary so much that a report which is right for one customer is quietly wrong for the next.
A connector isn't just a database connection. It's accumulated knowledge about how the application and the industry actually work. The report engine reads customer databases on SQL Server, MySQL, PostgreSQL, and Oracle, and most of the value lives in knowing what the data means.
Running it
Ops Insights runs on Kubernetes in AWS, defined entirely as code with OpenTofu, Helm, and Argo CD. Keeping it healthy has meant EKS version upgrades, autoscaling changes, a Redis-to-Valkey migration, and the kind of cross-layer debugging where a monitoring service fails to start because its storage volume landed in the wrong availability zone.
The application itself is a large, long-lived PHP codebase. I led its upgrade from PHP 7.4 through 8.3 to 8.5 in stages, using a code knowledge graph to trace dependencies before changing anything, so a decade of working behavior survived the move. The front end is now moving from jQuery and Bootstrap 3 to Vue, TypeScript, and Tailwind.