In our view the actual timetable solving (the maths part) should not be done with AI (especially in the meaning of LLMs) but with constraint optimization algorithms instead. LLMs are not that good at math and the timetabling problem is especially what contraint-optimization like CP-SAT is good at. So, in this part of the process AI does not help much but that is OK as it is not needed here.
But, AI and LLMs are extremely useful in couple of other use cases:
Most of the existing timetabling software tools are very settings-heavy and it takes a lot of time for the users and experts to click around different menus and settings to pre-design everything so the timetable that gets produces matches they requirements and preferences. A lot of this work can be simplified with AI. Instead of doing a lot of this work clicking in the user interface, we could feed school/planner preferences via natural language or via just dumping previous year approved timetables so the tool can extract those preferences and make settings adjustments based on those. For example by extracting certain time preferences for certain lessons and selecting those constraint weights that the school seems to prefer.
And instead of having the user to click around in a cluttered user interface, they could just tell the AI planner agent inside the tool to "move term 2 plan from this year into next year and adjust it taking into account national holidays" or something like that and the AI agent would then go and carry the task end-to-end which could take potentially even hours of users time if they would try do it manually.
