Healthcare Appointment System

A 100+ person daily queue against only 50–60 monthly tickets. The root cause was not demand — it was uncoordinated booking channels producing a thundering herd and race conditions on the same tickets.

A high-demand neurology centre where people queued physically and by phone for the same limited appointment slots. This was an analysis exercise, not a client production build.

• →Model the actual queue behaviour before proposing any technology.
• →Identify the contention point: multiple channels writing to the same ticket pool with no coordination.
• →Design for atomic assignment first, interface second.
• →Iterate — the first two designs were wrong, and the third was the one that accounted for the people the earlier ones forgot.

1. 01Two self-service channels: web and IVR
2. 02Atomic ticket assignment with TOCTOU protection
3. 03Hold-timer and booking TTL to release abandoned slots
4. 04IVR routing tree for non-web callers
5. 05One unified ticket database as the single source of truth

• System design
• Queueing analysis
• IVR flow design
• LLM-assisted design review

• +Treating the ticket pool as a concurrency problem rather than a UI problem.
• +Keeping an IVR channel so that non-web users were not excluded.
• +Using an AI model to find the person each design iteration forgot.

• −The first design assumed everyone would book on the web — it excluded a large part of the actual audience.
• −The second still allowed two channels to hand out the same ticket.

A documented architecture and a written breakdown of the three design iterations. Not a production system and no operational metrics are claimed.

System design starts with the wrong answer. The value is in noticing who the design excludes and iterating.

• Read the full write-up on dev.to →