The story

NurseList started with someone very close to me: my girlfriend, Tuğçe, an operating-room nurse.

I saw how much time and mental energy her weekly room and shift planning required. Nurses have to be assigned across different rooms and shifts while following many rules. Each room also has a different level of difficulty, so the workload needs to be shared fairly.

The plan may look manageable when everything stays the same, but a day-off request, a sick nurse, or an unavailable room can change everything. A small change can force the whole schedule to be reconsidered, sometimes from scratch.

Tuğçe was facing this problem in her daily work, so we decided to work on it together. We started with her experience, mapped the real constraints behind the printed plans and manual edits, and built NurseList as a simple, fast, and fair way to create and revise weekly operating-room schedules.

What I learned

The hardest part was not generating a schedule. It was defining the boundary between flexible language and reliable operations. GPT-5.6 is useful for understanding a nurse manager's written instruction, but a deterministic solver must remain responsible for Assignments, hard constraints, feasibility, and fairness. Human approval must remain the final decision.

I also learned that a fair schedule needs to be explainable. A single score is not enough; the user needs to understand which rule was preserved, which preference was relaxed, and why.

How I built it

NurseList is a Work & Productivity scheduling workspace for responsible nurses who need to create fair, explainable weekly operating-room schedules.

Manual planning has to balance leave, coverage, weekly shifts, room rotation, workload, and last-minute changes. NurseList turns a written operational instruction into a validated Planning Constraint with GPT-5.6, then uses a deterministic exact solver to produce a feasible Schedule Draft. The system exposes Fairness, Explained Exceptions, Assignment Swaps, and explicit human Schedule Approval instead of treating an AI-generated roster as authoritative.

The workflow is intentionally bounded:

1. Start from the approved Weekly Schedule and recent Schedule History.
2. Create the next Schedule Draft with one click.
3. Enter a plain-language change such as a leave request or room assignment.
4. GPT-5.6 interprets the instruction into validated Planning Constraints.
5. The deterministic planner recalculates the draft while preserving hard operational rules and optimizing a transparent mathematical cost model.
6. Review Fairness and Explained Exceptions, make a safe Assignment Swap if needed, and approve the draft.

GPT-5.6 has two meaningful roles in this project. Inside NurseList, it translates operational instructions into validated Planning Constraints; it does not create Assignments or approve schedules. During development through Codex, it helped me explore the repository, shape the domain model, implement the vertical workflow, write regression coverage, investigate release issues, and prepare the submission package. Human decisions retained the product scope, safety boundaries, and final approval responsibility.

The technical core

The technical core is an exact minimum-cost Assignment solver. Hard constraints such as Leave, Coverage Requirements, valid Shifts, and mandatory Planning Constraints must hold. Among feasible schedules, the solver minimizes assignment changes, Ideal Rotation deviations, missed Shift Alternation, repeated high-workload room assignments, and projected workload imbalance.

Challenges

The most challenging parts were handling changes without breaking the rest of the week, preserving fairness while satisfying mandatory coverage, and preventing ambiguous AI instructions from being partially applied. The product solves this by validating constraints before planning, keeping the solver deterministic, showing Explained Exceptions, and requiring explicit human approval.

NurseList was built during OpenAI Build Week with Codex and GPT-5.6. It uses representative data and is a focused demonstration, not a clinically certified or hospital-integrated production system.

Built With

• cloudflare-d1
• cloudflare-workers
• codex
• gpt-5.6
• react
• typescript