For an accurate low cost tiltmeter I suggest a Chinese machinists level bubble vial. The machinist's levels are good to 10 seconds when read visually. So optical sensing the bubble position with a photodiode array should get you to a fraction of an arc second.
http://www.shars.com/4-master-precisions-level-1The vials for machinist's levels are ground internally to a very large radius curve. The price is dependent upon how long the body is, so a short one saves money for your application.
However, as you are more concerned with change than accuracy, you could simply bend precision glass tube to make your own vials. The downside of that is that tilt in the cross direction leads to an error. So you would need to mount two on orthogonal axes. Use a slab of scrap stone kitchen counter top. Drill holes through the stone for clamps, lay the tubes flat on the stone and clamp the ends loosely. Slip a piece of of the thinnest shim stock you can find under the center of the tubes and tighten the clamps. The tube will deform to a uniform curve of large radius. You'll need precision shim stock of two different thicknesses which are as close to each other. Ideally you want 0.xxx0" and o.xxx5" material.
Connect the shim stock as one terminal in a capacitive bridge with two pieces of copper foil for the other terminal of the capacitors. Use something like a 5-10 MHz xtal clock chip as the signal for the capacitive bridge and attach a JFET input instrumentation amplifier to the copper foil pads. Adjust the position of the slab so that the amp outputs are close to zero. You'll need to hermetically seal the whole thing with a bag of silica gel and preferably a shot of dry nitrogen gas before closing it up. Or pull a hard vacuum on it.
You can, of course, also use optical sensing for homemade vials.
Have fun and let us know more about the project. I'm a geophysicist. I've never done any tiltmeter work, but am well aware of it. There are companies that can measure submillimeter deformations over miles using satellites or aircraft and laser or radar interferometry. It's pretty mind boggling stuff.
BTW another approach would be to use a differential pressure sensor with a liquid filled tube on each port. That's likely to be fragile though.