Engineering Deep-Dive: How Destm Built CuberiQ

Most headless CMS products stop at storage and a REST API.

Next.js

TypeScript

PostgreSQL

Redis

Claude API

Tailwind v4

Project details

Client: Destm internal — CuberiQ product team
Industry: Internal Product
Services: Product Engineering · Headless Content / CMS · AI / ML
Timeline: 14 months — v1 in production, ongoing
Year: 2026

The challenge

What we walked into.

Most headless CMS products stop at storage and a REST API. Editorial teams still hand-write meta tags, copy-paste between locales, beg engineers for form-to-API plumbing, and lose 30 to 60 minutes per landing page to mechanical work that should be automated.

We wanted a CMS where AI does the first 80% of authoring (drafting, translating, generating SEO and schema markup) and where business teams can wire forms to backend APIs — with auth, retries, and outcome handling — without a developer in the loop.

The risk was building yet another generic AI wrapper around a CMS. To earn the iQ-suite name we had to ship features that an experienced editor would feel daily — not a list of AI buzzwords on a marketing page.

What we built

The engineering decisions.

Stack — boring on purpose

Next.js + PostgreSQL + Redis + Tailwind v4. No exotic dependencies, no half-finished frameworks. Turbopack in dev, Vercel Edge in production. We chose boring infrastructure on purpose — the differentiation is in the product features, not the runtime.

Semantic content blocks

Every block knows its semantic role. A 'product highlight' block knows it needs a SKU reference. A 'testimonial' block knows it needs an attribution. The editor surfaces missing required references inline. This metadata also drives auto-SEO — JSON-LD product schema is generated from the block's semantic role, not from a separate authoring step.

AI authoring loop with the Claude API

Drafting, translation, and SEO meta use the Claude API with prompt caching tuned for editor-session reuse. Editors write a brief, AI drafts; editors edit; AI applies the edits as style fine-tuning for the next draft. We are explicit that AI does the first draft, humans approve — no auto-publish.

Forms-and-Actions runtime

Editors build a form in the visual editor, point its submit button at any backend HTTPS endpoint, and configure outcome states (success, partial fail, fail, auth error, network error) per form. Auth profiles live server-side; the browser only ever sees the public CuberiQ proxy URL — customer secrets never reach the page. This runtime now powers the destm.com /ai-audit lead flow.

Eating our own cooking

The destm.com website you are reading runs on CuberiQ. The AI Audit lead form posts through the Forms-and-Actions runtime. Every product launch, every blog post, and every landing page is the team using the editor in anger — which is the only credible way to find what is missing.

The results

Measured outcomes.

First-draft authoring time cut from ~45 min to ~7 min per landing page

SEO meta + JSON-LD schema generated automatically — zero engineer involvement per page

Locale variants shipped same-day instead of same-sprint

Forms-and-Actions runtime now powering Destm's own /ai-audit lead flow in production

The engineering behind this

Our ProductsThe iQ Suite this work feeds.AI & Deep LearningThe intelligence layer behind it.Data FoundationThe pipelines the platform runs on.

Next case study

Embedded Engineering for an Enterprise Luxury Fashion Brand

Have a similar challenge?

We'd love to hear about it. Every conversation starts with a senior engineer who's built systems like yours.

Let's Talk