Carsten Thomassen's article The Jordan-Schonflies Theorem and the Classification of Surfaces does exactly this: starting from the non-planarity of $K_{3,3}$, it proves the Jordan curve theorem.
Here is a part of the proof: this is Proposition 2.6 in the paper, and proves that a Jordan curve leaves the plane in at least two regions. Thomassen also proves at most two, but this is harder and not as self-contained. (Thomassen also proves the non-planarity of $K_{3,3}$ without the full power of the Jordan curve theorem, but I find the proof of Lemma 2.2 extremely unconvincing.)
Let $C$ be a simple closed curve in the plane. Choose:
- $\mathbf p$ to be a point on $C$ minimizing the $x$-coordinate;
- $\mathbf q$ to be a point on $C$ maximizing the $x$-coordinate.
Then $C$ consists of two paths from $\mathbf p$ to $\mathbf q$: call them $P_1$ and $P_2$.
For some intermediate value of $x$ between $\mathbf p_x$ and $\mathbf q_x$, draw a vertical line at that $x$-coordinate. Let $\mathbf r$ and $\mathbf s$ be points of that line lying on $P_1$ and $P_2$ respectively, such that the line segment $[\mathbf r, \mathbf s]$ does not intersect $C$. (The vertical line must intersect both $P_1$ and $P_2$ by the intermediate value theorem; if the intersection point with the highest $y$-coordinate is in $P_1$, choose $\mathbf s$ to be the intersection point in $P_2$ with the highest $y$-coordinate, and $\mathbf r$ to be the intersection point in $P_1$ above $\mathbf s$ with the lowest $y$-coordinate.)
We can find a $\mathbf p, \mathbf q$-path in the plane that does not intersect $C$ anywhere else: for instance, if we take a point $\mathbf a$ with $x$-coordinate lower and $y$-coordinate higher than any point of $C$, and a point $\mathbf b$ with $x$-coordinate higher and $y$-coordinate higher than any point of $C$, then the piecewise linear path from $\mathbf p$ to $\mathbf a$ to $\mathbf b$ to $\mathbf q$ cannot intersect $C$.
Finally, if $C$ does not divide the plane into two regions, pick a point on the $\mathbf p,\mathbf q$-path and a point on $[\mathbf r,\mathbf s]$, and draw a path connecting them that avoids $C$. Let $\mathbf t$ be the last point of the $\mathbf p, \mathbf q$-path on that path, and let $\mathbf u$ be the first point after $\mathbf t$ of the segment $[\mathbf r, \mathbf s]$ on that path.
Now we have a planar embedding of $K_{3,3}$ with vertices $\mathbf p, \mathbf q, \mathbf u$ in one part of the bipartition and vertices $\mathbf r, \mathbf s, \mathbf t$ in the other part.