ramine.net: AI-First Publishing

Live · started June 2026

Publish a body of work with its sources, working demonstrations and history intact.

A typed content graph, deterministic two-site compilation and agent-operable editorial workflows produce this site. Citations, search, media and fully prerendered routes come from the same records.

Visit product · Latest product decision

Scope: Research, product strategy, information architecture, design, engineering, infrastructure, QA, launch, and publishing operations

Open a project, try its demonstration, follow a claim to its source, or find a passage through search. This site keeps those experiences connected to the same authored records. The portfolio is also the working example of the publishing system I designed and built with AI agents.

The unusual part is not generating a page. It is letting a person or an agent change a body of work without quietly breaking its evidence, routes or publication boundaries. Typed content, deterministic builds, an editorial interface and output checks make those decisions inspectable. The result is live here: the case studies, recent decisions, citations and search are produced by that system.

One record, many reading paths

Projects, professional case studies and dated updates have different roles and explicit relationships. Citations resolve to durable evidence and number themselves within the document. Media passes through a reproducible pipeline for responsive images, posters and previews. Every public route has its own prerendered page and metadata, so a shared deep link does not depend on a client-side router rescuing it.

Search compiles titles, metadata, tags and full article text into a lazy in-memory index, without a runtime service or model call. The same content can also become the Spatial Gallery, an alternative way to explore the work rather than a separately maintained catalogue.

Publishing with agents without handing over authority

The local editorial interface edits and previews the same structured records that coding agents work on. Validation, fixtures, tests and reviewable Git commits keep both paths inside the same content and release boundaries. Agents help with research synthesis, implementation, media, testing and publishing operations; I decide what the site says, how it behaves and what is ready to release.

The compiler also produces a separate publication, CompanyCompany, with explicit content eligibility and its own identity, URLs, feeds and editorial presentation. Site selection happens at build time, never by guessing the runtime hostname. Output checks cover text, canonical metadata and reviewed media assets. These are deliberate publication boundaries, not a claim that the brands are anonymous or unlinkable.

What the system cannot decide

A valid record can still contain a weak claim. A passing build cannot prove that a native cursor looks right, a video plays correctly on every device, or an article is worth reading. Sources, attribution, interaction quality and editorial judgment still need review. The system makes that work visible and repeatable; it does not replace it.