I tried playing with it to [...]
This will be the same function as the accepted one above, but approached in a different way that might inspire how to get there by playing.
First, the question is really easy if the requirements are instead
- $b(-∞) = 0$
- $b(+∞) = 1$
- $b$ is monotonically increasing.
- $b$ is smooth.
There are a bunch of those, but let's pick $b(x) = ½(1+\tanh(½x))$.
Of course we didn't want an infinite domain, so we still have to compress $(-∞, +∞)$ down to $(0, 1)$.
What do we need for that?
- $u(0^{+}) = -∞$
- $u(1^{-}) = +∞$
- $u$ is monotonically increasing.
- $u$ is smooth.
Conveniently, that's also easy. We just add poles at $x = 0$ and $x = 1$, leaving us with
$$ u(x) = {1\over{1-x}} - {1\over{x}} $$
What if we combine them? https://www.desmos.com/calculator/oca6bndx7l

Which turns out to be exactly the same as the above answer:
$$\begin{align}
f(x) &= b(u(x)) \\
&= ½(1+\tanh(½u(x))) \\
&= ½{{2e^{u(x)}}\over{e^{u(x)}+1}} \\
&= {{1}\over{1+e^{-u(x)}}} \\
&= {{1}\over{1+e^{ {1\over{x}} - {1\over{1-x}} }}} \\
&= {{ e^{-{1}\over{x}} }\over{ e^{-{1}\over{x}} + e^{-{1}\over{1-x}} }} \\
&= {{ h(x) }\over{ h(x) + h(1-x) }}
\end{align}$$
It's left as an exercise to the reader to show why picking the logistic function for $b$ worked, but other things don't. For example, using arctan gives a discontinuous first derivative.