There is no way around it - at some point you're going to have to experiment with a leveling box. 

List all the products/deliveries that one cell produces

(In admin, all the jobs the team does for others)

Then split up the day/week/month in regular pitch intervals and, using Takt Time, draw cards for each slot where a work container/job is needed.

Whenever a container/job is picked up, use cards to reproduce what has just been sold in the same order, much like a restaurant does with orders from the tables. Don't overthink it or overcalculate it. Every restaurant in the world works in kanban, there is no reason your team can't.

And then run it until you see what are the strengths and weaknesses of the cell and start solving problems!

Should I pursue waste elimination or lead-time reduction?
If I turn off the MRP in favor of kanban, will the factory collapse around us?
What? My pull system is supposed to fail?
Why haven't kanban and value stream mapping improved delivery from a low-volume-high mix process?
Are you pulling?
Can a pull system with zero stock make the company more fragile?
Method vs. Tools?