A semantic AI engine that plugs into any CMS without touching it
An SEO SaaS founder needed his product to understand a site's pages well enough to link them automatically. We built the engine — and stayed on as engineering partners.
The problem
Internal linking is high-leverage SEO work that almost nobody does well at scale: it is manual, easy to get wrong, and the first thing that gets dropped under time pressure. The product had to crawl a site, understand what each page was actually about, and place internal links automatically — without the customer ever touching their CMS.
What we built
- A semantic suggestion engine: every page embedded via OpenAI, stored in per-site vector collections, link opportunities scored by meaning rather than keyword match.
- A zero-CMS-change injection script: a client-side script places approved links live, marks its own links to avoid re-injecting them, skips headings and embeds, and ships a diagnostic mode for verifying the install.
- A WordPress plugin and a Google Search Console integration, so link decisions are weighted by real performance data, not just similarity scores.
- A self-hostable, open-source alternative for customers who want the injection running on their own infrastructure — lock-in was never the point.
- A crawl pipeline with pause/resume, section exclusions, scheduled recrawls, credit metering and billing webhooks — the unglamorous half of running this as a product, not a demo.
- Later, an AI-run churn investigation across support tickets, product analytics and billing data through a custom internal tool we built — every AI claim cross-checked against the database before anyone acted on it.
Stack
Scale & result
Customers running the finished product have published their own results: one site saw 77% more organic clicks in six weeks, another saw 10× impressions in three months, a third tripled organic sessions in six months. For us, the bigger result was the engagement itself: it became a formal partnership.
More AI work we have shipped
Every claim above is a shipped, production system.
AI loan-intake chatbot
A bilingual chatbot that automated 80% of loan applications for a community-lending platform, cutting processing from days to minutes.
A film for every day’s news
MoviesEverywhere gets up to three film ideas a day for the news, each with a ready post. The agent learns from his votes. About $15 a month.
Tell us what you're building.
The problem, roughly what you think it needs, and where the data lives today. We will tell you what we would actually build — including when the answer is less than you asked for.