You correctly state that neutrinos do not interact too often. The physical parameter describing that is the effective cross-section. So what you observe in a detector is not the neutrino itself, but secondary particles, e.g. muons. Colloquially put, you may regard anything with high mass (density) in between the neutrino source and your instrument (to detect mainly the secondary particles) as detector for the neutrinos. This might be a large chunk of ice (like for the icecube experiment), a mountain range (as for the Gran Sasso lab), or even a complete planet, if the incident neutrinos originate from a star located on the other side of Earth at the moment of observation. The latter is possible since neutrinos e.g. from the Sun may pass through the whole planet without interacting with its matter (at night-time for the scientists working for the detector).
Let me try to explain with a drawing I took from a 2018 blog entry of the Antarctic Muon And Neutrino Detector Array (AMANDA). The red dots are on the ice surface, the vertical lines are the boreholes into the Antarctic ice shield and the little black dots are the PMTs, the Photomultiplier Tubes which detect the Cherenkov light generated by some fast muons, in the picture on the top of the blue cone. The readings of the PMTs (detecting the Cherenkov light) are displayed as color-code, I assume the color scale somehow relates to the time passed since the trigger event.

So how do we now know where the neutrino came from? We can use particle physics to derive the secondary particle's velocity (vector) and from that where the neutrino actually came from. This is not trivial and requires quite a bit of computer power, but it is possible.