Divide your drawing into three phases:
1) Draw all opaque 3D geometry. This uses the camera view and projection matrices, and the object world matrices.
2) Draw all blended 3D geometry. Sort it from far-to-near to the camera. This makes sure that windows etc properly blend in order. Again, use the object world and the camera view and projection matrices.
3) Draw all 2D/gui geometry. This uses screen space coordinates; the view/projection matrices com from the GUI system, not the camera.
In general, you can place something anywhere you want on the screen with the appropriate transform matrices. After the projection matrix, the screen coordinates are X=-1 for left, X=1 for right, Y=-1 for bottom and Y=1 for top. Z=0 is the "near" plane, and Z=1 is the "far" plane, after Z has been divided by W. If you like this coordinate system, you can even emit vertices directly in this coordinate system, and set all the transform matrices to identity (or use a shader that doesn't transform through matrices at all).
If you're unsure about how all this works, then I suggest reading up on
the 3D graphics transform pipeline -- it's pretty much required knowledge when working with 3D graphics.