Honestly i dont really like SA and PM (Scientific American and Popular Mechanic) because of the holes on the info they provide. It's like that Dilbert comment "Build your own helicopter from common household items, on PM!."
The first problem is the cell itself. I admit that small variations in wall thickness might not be a big issue, but when you increase the required accuracy, they could be. Just saying.
So having some guy from your local glass shop make one ... well. I'd go to a medical supply, or even scientific supply store and get one. I'd venture that it's not that expensive.
Another point is that glue thing. regular glue doesnt attach to glass very well, and neither for a long time. And using (as it stated) sand, which creates a very diffuse contact point, with glue, doesn't seem very stable, specially through temperature differences.
Third point is very simple. Any welding or soldering on sensor is something to avoid. Any weld made will change the ohmic value of the sensor, per wire. Add that to the ohmic value of the wire, which you should know, and size of each, which sould remain equal, and you get a mess to calibrate.
but hey, that's why labs exist
.
PS1: for that particular range, you really have to use a platinum resistance thermometer (PRT).
PS2: for the ones that haven't figured it out what a "cell" is supposed to be:
(10m, paint! -> best app ever)