The machinery
Most sites hide how they’re made. This one is the artifact — the data, the checks, and the tools are the work, so they are shown rather than described. Live demonstrations are mounted below; corpus claims come from a generated, machine-readable snapshot.
This is a coverage ledger, not a claim that every record has received the same depth of review. The build regenerates it from the checked-in data, and a unit test fails when the public snapshot falls behind.
Open the machine-readable snapshot · Read the source standard · See how claims are separated
Two of the projects behind this site run in the browser right now. They are embedded as they are — no screenshots, no mockups.
Scripture text comes from the WEB, KJV, Douay-Rheims, and Brenton's Septuagint. Build scripts produce versioned JSON for the reader, cross-references, topics, and source-language layers. A generated corpus snapshot and freshness test make drift visible.
Every chapter guide receives automated identity, reference, and quotation checks. New v2 batches also lock their grounding prompt, model, rubric, outline coverage, cited questions, and template-removal checks. Editorial review stays separate: agent review and named human sign-off are different published tiers.
Museum works record their institution, object page, and license. Generated interpretive plate manifests retain prompts, captions, and alt text instead of being presented as historical objects.
Lint, types, unit and smoke tests, a full static build, browser journeys, and rendered accessibility checks run in CI on every push to master. Exact test totals come from the runner rather than hand-written page copy.
The case studies walk through the projects one at a time — EYE, this site, and JobPilot. To discuss a project, start with how I work or write to garo.vartabed@gmail.com. The full-time employment profile remains available separately.