david_xu_sole

Ruloom: the idea and design Why I built Ruloom I built Ruloom to help freelance workers get useful practice data faster. Preparing a lesson or demo takes time finding and fixing files. I wanted to make that easier. One small teaching task Imagine teaching someone to follow one rock sample: find the drill-hole location on a map, read the depth, open the picture and look at the infrared chart. A small dataset is enough, but the files need to agree. From a request to useful files We start with a real example and suitable source material. Then we decide what the task needs. A model suggests a plan, code creates the records, and checks look for problems. A failed required check stops the release. How the parts fit together The engine keeps the steps in order. A domain profile adds rules for one kind of data. That lets us add another kind of data without rebuilding the whole process. What Muse contributes We’ve used Muse Spark to suggest settings and Muse Glimmer to review results. Next, we plan to give Muse models separate jobs: researching examples, planning the work, and reviewing data and pictures. Code checks the results, and a person decides whether to accept them. What we want to deliver The goal is a small ZIP with useful files, a guide and check results. That could reduce the need to share original records. We’re preparing a comparison with a separate Claude Code run on the same request and sources, to see whether someone can use each package.