Bhaktamar.org — five thousand pages, no database.

The material was never the problem. Jain devotional literature — hymns, verses, commentaries, ritual texts — exists in abundance, in print and in PDFs passed between families. What didn't exist was a way to find a specific line without already knowing which book it was in.

The audience makes that harder, not easier. Readers are mobile-first and skew older, often reading Devanagari at increased zoom on a mid-range Android phone. Many will search by typing a half-remembered line in Roman letters, because switching keyboards is friction.

So the work split in two: turn printed books into structured, validated records — and then build a search that meets people where they actually type.

The site was finished and deployed before it was ever submitted to a search engine. That was deliberate. A fresh domain's first crawl sets a quality baseline that is slow and expensive to move afterwards — so the fixing happened first, not after the traffic disappointed.

That sprint found real problems. Nine section pages were throwing a client-side exception in production, traced to structured data large enough to desynchronise the page's own streaming parser. Romanised search had been silently dead across the entire corpus because of a misplaced character in a build script. The whole content library was leaking into the browser bundle and pushing the build to twelve gigabytes of memory.

None of those would have shown up in a page that merely looked correct. They surfaced because the build was made to check itself.