These are the two ways I'm trying to choose one of. The sketches should help understanding how the sensor is going to work.

The sensor will rely on the red-eye effect: the light of a LED will form an image on the retina. The backscattering from retina will "send" the image of the LED back  into the actual LED. One can sense that either by placing a photodiode really really close to the LED (on config.1), or using a semitransparent mirror to split the beams (on config.2). The light reaching photodiode will be at maximum when the image of LED/photodiode is exactly on the retina, i.e. is precisely in focus.