I'm sure I'm going to miss some mathematical detail in this, but I want to make sure that I have the gist of the $\mathfrak{so}(1,3) \to \mathfrak{su}(2)\oplus\mathfrak{su}(2)$ relationship correct.
$\mathfrak{so}(1,3)$ is formed of the six generators of proper orthochronous Lorentz group, $\mathrm{SO}(1,3)$, three $J_{i}$ and three $K_{i}$, with the Lie brackets:
$$ [J_i,J_j] = i\varepsilon_{ijk}J_k \\ [K_i,K_j] = -i\varepsilon_{ijk}J_k \\ [J_i,K_j] = i\varepsilon_{ijk}K_k $$
If you redefine these as $A_p = \frac{1}{2}(J_p + iK_p)$ and $B_p = \frac{1}{2}(J_p - iK_p)$, these new generators span a subset of complexification of $\mathfrak{so}(1,3)$, $\mathfrak{so}(1,3)_{\mathbb{C}}$, that is in fact isomorphic to $\mathfrak{so}(1,3)$.
The new Lie brackets are then given by:
$$ [A_p,A_q] = i\varepsilon_{pqr}A_r \\ [B_p,B_q] = i\varepsilon_{pqr}B_r \\ [A_p,B_q] = 0 $$
so in fact the $\{A_p\}$ are isomorphic to $\mathfrak{su}(2)$, as are the $\{B_p\}$ independently. Hence $\mathfrak{so}(1,3)$ is isomorphic to a restriction of the full $\mathfrak{su}(2)\oplus\mathfrak{su}(2)$ Lie algebra to just generators of the form:
$$ A_p = \sigma_p \otimes I \\ B_p = I \otimes \sigma_p $$
Then, by rearranging $A_p = \frac{1}{2}(J_p + iK_p)$ and $B_p = \frac{1}{2}(J_p - iK_p)$ we get:
$$ J_p = A_p + B_p \\ K_p = -i(A_p - B_p) \\ $$
from which it follows that:
$$ J_p = \sigma_p \otimes I + I \otimes \sigma_p \\ K_p = -i(\sigma_p \otimes I - I \otimes \sigma_p) $$
which is the standard result that's on the Wikipedia page (barring the extra minus sign I seem to have in $K_p$...).
Hence, you can find the representations of $J_p$ and $K_p$ in terms of the representations of $\mathfrak{su}(2)$, which are well-known. What I don't understand is why you have the freedom to choose different dimensions for the representations of the two copies of $\mathfrak{su}(2)$. Is it just because the tensor product preserves the dimension of the product of square matrices under commutation?