Peakseek
Genetic algorithms as a service — an evolution loop you run against your own fitness function.
The Loop
How it works
- We suggest an organism to test — drawn at random from the organisms you haven't scored yet.
- You test it on your infrastructure, against your fitness.
- You report one number: the fitness you measured.
- We breed offspring — crossover and mutation — from the outcomes.
- Repeat: every generation gets more informative.
Your fitness
Your fitness function stays yours
We never run or evaluate your fitness function. You test every suggestion on your own infrastructure, and the only number we see is the one you report.
Pricing posture
Pricing
Free tier — the basic loop
Suggestions, one fitness number per suggestion, and offspring breeding. Free while we learn who really needs this.
A paid tier is planned for later. Nothing about it is built or promised yet.
What is a genetic algorithm?
An evolution loop that searches your design space
A genetic algorithm is an evolutionary loop: we vary candidate solutions, you test them, the best ones get selected, and we breed the next round. Variation → test → selection → repeat. Each cycle searches a design space for candidates that score well against a fitness function — the one number per candidate that measures what you actually care about.
The fitness function stays yours. We never run or evaluate it — you test every suggestion on your own infrastructure and report one number back.
Use cases
Where one number per candidate is enough
Payment form tips
A tip amount is a genome: three amounts you tune. We suggest a configuration, your checkout measures the one number that matters, and we breed the next generation of tip suggestions.
Design parameter search
Any product with a recipe — layout, thresholds, component sizing — where each candidate can be scored. We propose, you measure on your box, we evolve toward better candidates.
Hyperparameter selection
Pick a set of numeric knobs for a model or process, score each combination once, and let the loop find the region worth exploring next — one reported number per trial.
Create a genome
From blank chromosome to first suggestion
A genome is a set of typed alleles that define the value space you want to search. The designer lets you build one with four allele types: Float, Integer, Boolean, and Option. Set a population size and evolution thresholds, turn it into an experiment, and request your first suggestion.
