How Cerbo Compressed a 3-Year Modernization to 9 Months with AI-Accelerated Engineering

The Challenge

Cerbo builds clinic and hospital management software that physicians use to track patient records, medications, and medical histories. As the product grew, so did the codebase, but without a methodology to match. Each new client request meant copying a chunk of code, adding features, and deploying a new version. Over time, that pattern produced roughly 2,000 divergent versions of the same software. The result was a codebase that was functionally impossible to maintain, nearly impossible to audit, and a significant obstacle to any new product development.

Traditional approaches to resolving this kind of technical debt (large teams of developers, multi-year timelines, significant capital outlay) were on the table. But the economics were hard to justify, and the risk of disrupting a production system in active clinical use made the conventional path even less attractive.

→ See how Sparq approaches production-critical AI deployments through The Shop, where every engagement is stress-tested under real operational load before it touches a client's systems.

The Solution

Sparq brought an AI-accelerated engineering approach to an engagement that, by any traditional estimate, would have required a large team and several years to complete. Working alongside Cerbo's technical lead, Sparq built the tooling, methodology, and documentation infrastructure needed to make AI effective on a long-horizon, production-critical project.

A centerpiece of the technical approach was PHP Similar, a custom command-line tool built by Sparq in two days. It functions like a PHP compiler, but outputs an abstract representation of each file, stripped of variable names, comments, and whitespace. That abstraction layer dramatically reduced false positives in duplicate detection and gave the AI a reliable, trustworthy mechanism for making data-analysis decisions. Specifically, whether a file could be safely removed.

Sparq also engineered a context management system to address one of the core challenges of using AI on long-term projects: statelessness. Every ticket analyzed, every decision made, and every failed attempt were documented in a structured markdown directory within the IDE. That documentation keeps the AI grounded across sessions and provides a full decision history that could support team scaling if the engagement expands.

Testing ran in parallel throughout. Each new implementation followed a test-driven development cycle: plan first, build test cases second, code third. End-to-end validation ran through Playwright (UI and API), Docker containers, and GitHub Actions before any changes hit production—a layered approach that kept AI-induced errors to a minimum throughout the engagement.

→ Sparq's Intelligence Studio is the embedded AI delivery layer that makes this approach deployable across industries, injecting intelligence into existing systems without replacing core infrastructure.

Outcomes

Sparq completed approximately 35% of the total modernization scope in the first three months of engagement. 54,000 files have been deleted from production. Over 22,000 duplicate security profile files were identified and removed. In a single month, more than 1,200 files were deduplicated, a milestone the team described as a significant inflection point in the engagement's pace.

The effort that would have required a large team working for 2-3 years under significant operational pressure is now tracking to complete all three phases of the modernization, including new feature development, by the end of 2026.

Phase one, the deletion and deduplication of redundant code while preserving full functionality, is well underway. Phase two addresses straightforward configurations. Phase three expands the software itself.

The engagement is ongoing. When it closes, Cerbo will have a maintainable, extensible codebase capable of supporting product development that its technical debt had made structurally unavailable for years.

→ If you're in or approaching a legacy modernization engagement, the architecture decisions being made now determine what's possible in 2027 and beyond. See what modernization built for operational capability looks like.