A content engine that runs itself
My role
A team build at Cognilium (lead engineer: Ali Ahmed). My owned areas in 2026: research, the SEO/GEO/News-SEO engine, the content engine, cross-platform syndication, and the build documentation.
The problem
Cognilium's main site needed to win traditional search and AI answer engines while continuing to publish without manual SEO work on every post.
What I did
- Owned the site surface and the content engine: 76 routes, a 22-post blog organized into 6 pillar-and-spoke topic clusters, a tech-news newsroom, and a GraphRAG pillar page plus AI Engineering Glossary.
- Built on Sanity (GROQ, Portable Text, image CDN) with webhook-driven ISR, so editors publish and the site revalidates exactly the right pages automatically.
- Ran the Strapi → Sanity and SendGrid → Resend migrations, ripping out the old CMS and rebuilding the newsletter as proper double-opt-in.
- Wrote 90KB+ of stack, platform, and growth documentation the team onboards from, plus the architecture and workflow diagrams.
The result
A live company site and content engine: publish once in the CMS and it is structured, optimized, and indexed automatically, with the SEO, News-SEO, and distribution systems layered on top.
The judgment call: what the AI couldn't do
I made the schema system throw at build time when a required field is missing, so a malformed or schema-less page literally cannot deploy. The AI will happily ship a page with no Organization markup; the build refuses it. Encoding the standard as a build gate is what keeps quality from sliding as volume grows.
By the numbers
0
routes live
0
components
0
blog posts / 6 clusters
0
platform migrations
Proof
Verify: View-source the schema, RSS feeds, and llms.txt.