Short answer
Do you find the null position, then assume it's 90 degrees from the
beacon?
That's correct. For the antenna pattern shown in the question, the angle between direction of nulls and peaks is 90°. When sensing a null (or a peak) there are two possible and opposite directions for the beacon. This ambiguity has to be removed. This could be done by triangulation, with a second reading of the angle of arrival after the receiver has moved a bit, but in the modern ADF antenna this is not required, a sense antenna is added to the loop antenna to change the radiation pattern (i.e. the sensitivity vs. the azimuth):

Radiation pattern of the loop antenna, and loop + sense antenna system (measured in a horizontal plane)
With the sense antenna, the null is at 180° of the peak, and there is no ambiguity. However the performance of the system is degraded by the sense antenna, thus usually two measures are done, without the sense antenna (for maximum accuracy) and with the sense antenna (for ambiguity removal only). See further down for the details.
The reason we prefer to sense the null direction is the signal received is fading at the larger rate near the null than it is increasing in intensity near the peak. The change being more pronounced near the null, sensing the null is easier and gives a better accuracy.
Electromagnetic waves simply
Contrary to what is often thought, the classical high-level principles behind EM waves, which explain how the loop antenna can work, are quite easy to summarize.
The very common representation of an EM wave:

Electric and magnetic components of an EM wave. Source
A wave propagates in 3D space and its electric and magnetic fields are not dissociable, they are a single electromagnetic field which energy is carried by photons (though according to Richard Feynman, EM sources produce not physical particles or waves, but wave-like “probability amplitudes” that propagate at c in space. QED! Well... let's stick to good ol' photons!).
A NDB transmitter is an electric generator, generating an alternating current in a basic vertical conductor which length is tweaked to make it resonant at the frequency used (it's a dipole antenna). Three laws are then governing what happens:
Ampère-Maxwell's law says 1/ the vertical current creates a circular magnetic field around the antenna (Ampère's initial idea). The field intensity follows the variation of the current, 2/ the same happens with an electric field instead of a current (Maxwell addition).
Faraday's law of induction says a varying magnetic field creates an electric field. This field opposes the magnetic flux which created it according to Lenz's law. Lenz's law is a kind of reaction principle.
So AC current creates a variable magnetic field. The energy of the magnetic field comes from the current in the antenna.
As the magnetic field is variable, using the two laws above alternately, we see when a magnetic field varies a variable electric field is generated. When an electric field varies a variable magnetic field is generated.
The energy of the electric field comes from the magnetic field, as postulated in Lenz's law: The electric field opposes the magnetic field changes, therefore it is a force, aka the back electromotive force (back-emf).
What's great is an actual current (electrons) is required only to generate the initial magnetic field, the subsequent magnetic fields are due to electric fields, not depending on a conductor and/or electrons to expand:

Electric and magnetic fields generated from a point of a vertical antenna in a given direction
From left to right: A current is created in the transmitter antenna. If we focus on any point of the antenna, the local current creates a magnetic field around the point. This field induces now vertical loops of electric field, each loop in turn creates a new magnetic field, and so on.
The two fields, normal to each other, are the two inseparable aspects of the electromagnetic field. They exist at the same time, and the repeating process leads to their propagation as wave. As fields have no mass, they can travel at the speed of light.
This process occurs in 3D space, we see concentric shells of electric field radiating from the whole length of the antenna, and between these shells the magnetic field linking the shells:

Still image of electric and magnetic fields. Source (animation).
The electric field at the top and bottom of the antenna is missing, we can predict a "cone of silence" at the vertical of some radio aids using vertical antennas.
The wave crosses conductors while expanding. A conductor acts as a reception antenna. Depending on the conductor ability to couple more or less strongly with the electric and/or magnetic fields, a given portion of the energy from the fields is converted back into current, by virtue of the same reversible laws.
ADF principle
The ADF receiver uses a small-loop antenna (the circumference is small compared to the wavelength), which principle is to sense the magnetic field of the wave, the antenna is actually a coil.
This loop antenna is unusual in the sense receiving antennas generally couple with the electric field of the wave. However sensing the magnetic field is efficient at lowest frequencies and for direction finding applications.
How is the null position created? Wouldn't the signal still be
received at that position, how does it become null?
Correct, the signal reaches the antenna with the same mean intensity, regardless of the orientation.
Let's look at an everyday life example: Hearing a sound without seeing the source doesn't prevent us to determine its direction. Unless the sound is coming from straight ahead or straight behind, one ear receives the sound first, the same for any reflection, our brain does the rest. Moreover our ear pinna is not symmetrical, this also allows to separate a front sound from a back sound at the same distance, or one low from one high. This localization is irrespective of the mean intensity of the sounds.
Likewise the loop antenna reacts to the combined result of instantaneous amplitudes at all points of the antenna. According to the law of induction, two points of the loop receiving the signal with a phase difference due to the difference of distance from the transmitter (left side below) will be at a different potential, creating a current in the conductor between these points:

ADF antenna sensing NDB magnetic field (not to scale)
On the other hand (right side), when the loop plane is normal to the direction of the wave, each point of the loop receives the magnetic field with the same intensity, no back-emf is created, no difference of potential is measured, no current circulates, this is a null.
- Note the antenna is actually much smaller than the wavelength and the phase difference is very small. But the loop antenna is a coil, it is easy to have multiple turns of wire, each turn "collecting" an additional amount of the magnetic field. However as the increasing impedance cancels partly this gain, a reasonable compromise has to be found.
"LW/SW receiver" with a ferrite coil antenna (a coil which inductance is reinforced by a ferrite rod) must be oriented in the direction of the station to maximize reception.
Rate of variation of gain
We mentioned earlier the radiation pattern of an antenna, which is a representation of its relative gain according to the direction. For the loop antenna:

Radiation pattern of the loop antenna (dotted line for the vertical plane). Source
The gain is provided in decibels, which is the logarithm of the ratio between the intensity in the direction considered, and the intensity in the direction of the maximum. By principle the maximum is 0dB (100%) and other values are negative (less than 100%).
When looking at the indications in green, The dip in received energy (the null) near +/- 90° is narrower than the peak near 0°/180°. This tells us about the rate of variation of the sensitivity:
Signal is maximum at 0°, but varies only by less than 2 dB (1.2x) on a range of 60° around the maximum.
Signal varies by more than 30 dB (31x) for only 30° around the minimum/null.
It means it will be more accurate to determine the direction of the null rather than the direction of the peak (which is the direction of the transmitter). This is the case of many types of antenna: Nulls are more pronounced than maximums.
Loop antenna with sense antenna
There is a problem with the loop antenna: It has a symmetrical radiation pattern, when a null is found, there are still two possible opposite directions for the transmitter. The solution is to use an additional element, the sense antenna, which is an antenna with the same sensitivity regardless of the azimuth. It is said to be omnidirectional and is usually usually coupling with the electric field. The two antennas can be combined in such a way the resulting pattern has a heart shape (cardioid):

Radiation pattern of the loop antenna combined with sense antenna
One lobe of the loop antenna (here in blue) adds with the sense antenna, the other is subtracted from the sense antenna, creating a gain dissymmetry.
The ambiguity isn't possible anymore. The null is now opposite to the maximum, while they where at 90° before adding the sense element. When the null direction has been determined, the NDB direction is the exact opposite.
Back to the audio field, this antenna system is very similar to the sound engineering technique called MS stereo using one mono (mid) and one stereo (sides) microphones, though this technique is valid for only a 180° azimuth range.
The usual procedure to determine the direction is:
First find a null with the sense antenna off, because the sense antenna smooths out the null dip in the radiation pattern. This give the direction with 180° ambiguity.
Do a second but rough measure with the sense antenna on to remove the ambiguity.
Modern ADF antenna
On modern ADF, the antenna doesn't rotate to determine the direction of the NDB. Instead the antenna system is a combination of multiple fixed antennas. The signal direction of arrival is determined by sensing the time of arrival at the different individual antennas. As the time of arrival is actually difficult to determine without an accurate clock, the difference of signal phase is used, as explained in this other question:
Since modern ADF receivers don't rotate, how do they work?
A practical realization:

Source