Directed Altitude's company-wide pivot from rule-based decision engines to LLM- and agent-driven pipelines for clinical intelligence. Architected agent orchestration, replaced the legacy rules engine with an LLM-powered reasoning system, and built the evaluation and governance framework that lets clinical AI ship safely.
What was broken
Altitude's clinical decision support ran on a hand-coded rules engine. Every new condition or workflow meant another branch in the rule tree, another regression to test, another conversation with the clinical team about edge cases. The system was reliable in narrow lanes and brittle everywhere else.
The team also needed to extend into new conditions and workflows quickly to win enterprise pilots — and the rules-engine substrate was the wrong tool for that pace.
What we built
We designed an agent-orchestrated pipeline where reasoning is performed by an LLM grounded in clinical context, with deterministic retrieval and policy layers around it. Skill files encode the playbook for each clinical scenario; an evaluation harness with golden cases gates every change.
We integrated with Athenahealth, went through Marketplace approval, and shipped to production. Clinical AI outputs are reviewed through a governance framework that tracks drift, flags hallucination signals, and routes ambiguous cases to humans by design.
Why it matters
The rules engine wasn't replaced because LLMs are exciting. It was replaced because the new substrate lets clinicians extend the system in their own language, lets the company ship workflows in days instead of quarters, and lets the safety story be auditable rather than asserted.