The Problem
A hospital was managing patient-therapist scheduling through spreadsheets. A common approach, but one that doesn't scale: as patient volume grows, matching the right therapist to the right patient at the right time becomes increasingly complex. I asked: could AI take this over? The short answer was yes, but not alone.
Early AI models like ChatGPT are powerful reasoning partners, but their outputs are probabilistic. Ask the same scheduling question twice and you might get two different answers. For healthcare scheduling, where patient safety and provider availability are non-negotiable, that isn't acceptable.
The Approach
Rather than forcing AI to be something it isn't, I designed a hybrid system: AI handles the reasoning and rule articulation, a deterministic rule engine handles the execution.
Working with ChatGPT, I built the system layer by layer. First, data schemas for patients and providers using synthetic data to protect privacy. Then, scheduling rules capturing real-world constraints: provider specialties, patient needs, availability windows, and locked appointments that couldn't be moved. Finally, validator scripts to verify the outputs met every rule before confirming a schedule.
The AI never scheduled directly. It designed the logic. The rule engine executed it.
The Outcome
The result was a near-deterministic scheduling system that consistently matched patients to the right therapist, at the right time, within established constraints. What previously took manual effort across multiple spreadsheets now ran as a repeatable, auditable process.
The Bigger Picture
Any organization still managing scheduling through spreadsheets is sitting on an opportunity. The combination of AI reasoning and rule-based execution is not complex to build, but the impact is immediate: fewer errors, faster scheduling, and a process that scales without adding headcount.
This experiment was built through prompting alone, before I discovered coding agents like Claude Code. That constraint turned out to be a strength: it forced a clear separation between what AI should decide and what rules should enforce. That principle holds today regardless of which AI tool you use.