Skip to content
WORDPRESS AUTOMATION · CLIENT Fine Touch Marketing

SEO Operations

AI metadata jobs with validation, per-item results and recovery.

Illustration of SEO Operations: a metadata job table with per-item validation results and worker health
ILLUSTRATION
CURRENT SCOPE

Operational PHP/Python workflow

WHO IT'S FOR

Client project · Fine Touch Marketing

Med-spa marketing agency. SEO metadata automation across their clients' WordPress sites.

THE ENGINEERING

A protected operator interface queues WordPress metadata work through a PHP/Python runtime. Title normalization, industry and location constraints, deterministic repair and per-item results govern AIOSEO updates. September recovery work added supervised restart and boot recovery, separate frontend/API health checks and fault-injection verification. Twelve scoped recovery criteria passed in the retained record. Later headline, quality and article features were deployed, but their paid end-to-end generation test still needs a usable WordPress test target. The historical n8n implementation is separate from the current runtime.

FlaskWordPressAsync jobs

BEFORE → AFTER

Real problems from this build, what I changed and the verified result.

  1. CASE 01
    Before

    Interface availability and background-worker health are separate operational signals.

    After

    Twelve scoped recovery criteria passed in the retained September record.

    How I fixed it

    Added separate frontend/API checks and supervised restart and boot recovery.

  2. CASE 02
    Before

    Titles failed to match because the site and the operator encoded the same characters differently.

    After

    Updates land on the intended page instead of silently skipping.

    How I fixed it

    Normalized both sides with one shared contract covering entities, punctuation and dash variants.

  3. CASE 03
    Before

    AI examples leaked the wrong business context, and output sometimes hardcoded a city.

    After

    Copy stays on-industry and location-correct, checked per item.

    How I fixed it

    Locked the prompt to the supplied industry, derived the city per page, and added deterministic repair after generation.

Evidence reviewed: September 4 recovery records · October 10 public case study