It's on GitHub, fork it and send a pull request. Always send pull requests sober.
Change some ingredients and A/B test them.
With initial recipes like these, you'd quickly reach very low local maxima.
git history serves as a Tabu list, and different people's tastes will serve as a stochastic breakout. If lots of people have a go it will de-facto be a particle swarm tabu. Even with a single starting point there's a good chance of global optimisation.
Yeah, when I saw jmp I thought it looked impressive, but just reading the ingredients made me want to Ralph.