Another perspective on this question: how we can also get the sine and cosine addition formulae from this construction.
Gelfand's method is very clever: splitting a ratio of unrelated lengths into multiplying two ratios, each of which is a trigonometric ratio. However, I do not find this very natural to attempt at first glance.
More natural would be to use the line segments $BQ$ and $AD$, and their parallel segments $CD$ and $PC$. Using this we can observe that:
$$AQ + QD = AD \implies AQ + PC = AD$$
$$BP + PQ = BQ \implies BP + CD = BQ$$
Now as $AD$ is parallel to $PC$, $\angle PCA = \alpha - \beta \implies \angle PBC = \alpha - \beta$, so $\Delta ACD \sim \angle \Delta BCP$. Thus:
$$AB \cos \alpha + BC \sin(\alpha - \beta) = AC \cos(\alpha - \beta) \tag{1}$$
$$BC \cos(\alpha - \beta) + AC \sin(\alpha - \beta) = AB \sin \alpha \tag{2}$$
and as $BC = AB \sin \beta, AC = AB \cos \beta$:
$$\cos \alpha + \sin \beta \sin(\alpha - \beta) = \cos \beta \cos(\alpha - \beta) \tag{3}$$
$$\sin \beta \cos(\alpha - \beta) + \cos \beta \sin(\alpha - \beta) = \sin \alpha \tag{4}$$
These formulae look awfully close to what we want to prove. So indeed, relabelling $\alpha - \beta = \alpha', \beta = \beta'$, we have $\alpha = \alpha' + \beta'$ and making $\cos \alpha$ the subject in equation $(3)$:
$$\cos(\alpha' + \beta') = \cos \beta' \cos \alpha' - \sin \beta' \sin \alpha'$$
$$\sin(\alpha' + \beta') = \sin \beta' \cos \alpha' + \cos \beta' \sin \alpha'$$
which yield the cosine and sine addition formulae.