Scheduling of agents to tasks, in this case, activities to teams, is a combinatorial optimisation problem.
I solved this problem by modelling it as a directed acyclic graph, then using iteratively deepening A* search to search the solution space for an optimal solution.
I used heuristics to help guide the search in the right direction. Schedules, in which the lunch or presentation times didn't align with the rules, scored badly in the "evaluate" function in the Schedule class. This gave them a high cost which meant the algorithm would explore other paths (that would have hopefully) better prospects.
$ git clone https://github.com/JamesArthurHolland/DeloitteSchedule
$ cd src/com/company
$ javac *.java && java -classpath ../../ com.company.Main