brainjs/ucb-js

Imagine you have three different headlines for your website and you want to figure out which one gets the most clicks. Instead of running a traditional A/B test where you split traffic evenly and wait weeks for results, you could use a "multi-armed bandit" algorithm. These algorithms dynamically shift traffic toward the best-performing option as data comes in, saving time and avoiding lost opportunities. This library brings that optimization technique to JavaScript. The specific approach used here is called "upper confidence bound." At a high level, it balances two competing goals: showing the option that has performed well so far (exploiting what works) and occasionally testing the other options to make sure you're not missing something better (exploring the unknown). It handles this math automatically, giving more chances to proven winners while still leaving room for underdogs to catch up if they start performing better. You would use this if you are building a web application that needs to optimize user choices in real time. For example, an e-commerce founder could use it to test different product images on a landing page, or a PM could apply it to dynamically pick which onboarding flow to show new signups. Instead of manually analyzing results and flipping traffic splits, you let the library handle the ongoing decisions. The workflow is straightforward: you tell it how many options exist, it tells you which one to show next, and then you report back how well it performed (the reward). It also includes a serialize feature, meaning you can save the current state of your experiment to a database or file and pick up right where you left off after a server restart. The project is written in TypeScript and runs on Node.js.