SEO Wander
An internal automation platform where every destination is a plugin, and a broken one is data.
- Type
- Internal tool
- Industry
- SEO / marketing tech
- Platform
- Internal web application
- Role
- Design, build, deploy
- Status
- Built for our own use, not publicly listed
What made this hard.
An internal SEO automation platform we built for our own use: plan a campaign, generate content with an LLM, and submit it to a fleet of destinations: paste sites, blog platforms, press wires, directories, social bookmarking and search-engine index notification.
Every destination has its own form, its own quirks and its own way of breaking. Hard-coding them produces a codebase where one site's redesign is an outage for all of them.
Content generation is also slow and bursty, so none of it can happen inside a request, and the moment a platform starts acting on its own conclusions, a wrong conclusion becomes an automated mistake at scale.
What we built.
- Built a plugin architecture: an abstract base-platform class with shared browser-automation helpers plus a registry, so each destination is a self-contained handler. Handlers span paste sites, blog platforms, press-release wires, business and article directories, social bookmarking, and IndexNow-style index notification.
- Gave every handler an explicit is functional flag, so a destination that breaks is disabled as data rather than deleted as code, and can be re-enabled when it recovers.
- Made target selection adaptive: platforms accumulate observed success history, and a target tried enough times with poor results is penalised, more strongly at higher confidence.
- Kept reporting recommendation-first and non-destructive. An endpoint surfaces retryable failures, link stripping, degraded targets, account blockers and pause candidates for a human to act on, instead of auto-applying changes.
- Ran cross-cutting correctness as documented sweeps across the whole handler fleet: normalising parameter names, converting HTML to markdown before writing into plain-text fields, trailing-slash normalisation in link injection, published-URL verification with retry, duplicate guards, CAPTCHA detection and typed error classification.
- Wrote the conventions down as rules future handlers must follow (which parameter to read, always strip HTML before plain-text fields), because that is what keeps a large handler fleet consistent.
What it runs on.
Chosen to be maintainable for years, not to win an argument on the internet.
- Python
- FastAPI
- SQLAlchemy
- Alembic
- Celery
- Redis
- React
- Vite
- Stripe
Where it landed.
Long-running work runs on Celery workers and beat rather than in a request, so generation and submission can take as long as they need without a user watching a spinner.
This is the oldest project in the set, so treat the stack description as point-in-time. The architecture is the part worth borrowing.
Need something in this territory? This is AI agent & LLM development work. Start from all services or see pricing.
Related case studies.
Your project
Tell us what you're building. We reply within two business days.