build log: jan 31 — apprentice takes shape
this build log is automatically generated
what i shipped today
i shipped a ton of new features for apprentice, my ai-powered art study app. that includes user profiles, curated lists, and the ability to upload your own artwork for side-by-side comparison. plus, i finally started thinking seriously about security.
the wikiart dataset is… interesting
i’m building apprentice to help people learn art, and that means a curated dataset. i started with wikiart, figuring it would be comprehensive. turns out, it’s got some serious gaps. not just in terms of artist representation, but also in beginner-friendly content.
that meant writing a filter script to bias results toward works that are clear, well-lit, and not too complex. it’s a reminder that data is never neutral. it always reflects the biases of its creators.
claude code is convenient, but…
i used claude code to build a production chatgpt app. and in the back of my head i couldn’t help but ask… is it secure?
that’s why i spent some time thinking about the security risks of using llms for code generation. things like secret leakage and prompt injection. the dangerous part: you might not notice. the output looks like normal code. the commit looks routine. but the secret is now in your git history or sent to an external server.
i’m now using .claudeignore to exclude sensitive files from claude’s context. new rule… never rely on local .env files for secret management.
balancing building and shipping
it’s tempting to keep building features for apprentice. adding more artworks, refining the ai, and making it perfect. but the real cost is the attention it pulls toward building. i need to start thinking about distribution. that means a landing page, legal pages, and maybe even a full saas version.
but for now, i’m focusing on the core experience. getting the learning loop right.
david crowe — reducibl.com