You should have focused manually on the background stars. You then have to magnify the picture and then focus on a star, the star should change from a disk to just a point when optimal focus is achieved. This would have improved the picture, but the depth of field may not have been sufficient to get the house in the foreground in good focus. A smaller aperture of, say, F/8 would have been better, that would also have reduced the blur due to lens imperfections while you would not yet have had any diffraction blur. The ISO value would have had to be chosen larger, say, 16,000. To reduce the noise to acceptable levels, you could then have taken, say, ten pictures and then using image stacking methods, aligned the pictures and then the average would have been a low noise image.

Now, using the picture you took, it is possible to correct for the unsharpness using deconvolution by using the stars as the point spread function. I took an average of 6 stars to calculate the point spread functions and used that to deconvolve the image:

This should ideally be done on the original image, not the jpeg version as that causes the jpeg artifacts to be enhanced as well. Also, the image was likely resized, no star trails were visible despite the 30 seconds exposure. If you want to do deconvolution on the original image, then you cannot use the stars as the point spread functions, although you could first deconvolve images of the stars using line segments with the length of the trails as point spread functions to get to the desired point spread function with which you can deconvolve the rest of the image.

To avoid getting star trails in the image requires using a shorter exposure time, at a focal length of 17 mm, you can expose for about 6 seconds.