Context
The client operates two long-running US publications serving fire service professionals, with decades of editorial archive and an active daily publishing cadence. Their WordPress estate had grown organically: search was keyword-only and surfaced shallow results, editorial workflows leaned on manual repetition, and the AI capabilities their audience was starting to expect - summaries, semantic discovery, content automation - sat outside what off-the-shelf plugins could deliver without locking the business into a single vendor.
Brief
Modernise the publishing stack with production-grade AI capabilities, without committing to a single AI provider, and without rebuilding the underlying WordPress estate.
Approach
Rather than wiring features directly to one AI vendor, the foundation is a custom provider abstraction layer that supports Claude, GPT, Gemini, DeepSeek, OpenRouter, LiteLLM, and self-hosted Ollama through a single interface. BYOK throughout - keys stay with the client, providers are hot-swappable, and failover is automatic. Every feature built on top inherits that freedom.
Seven plugins now run in production across the publication brands:
- AI Suite (core) - provider abstraction, usage tracking, addon framework
- AI Summaries - article summarisation, AEO meta descriptions, bullet-point blocks
- Elasticsearch - hybrid keyword/semantic search using ELSER and ML reranking
- External Media Storage - S3/R2/Wasabi/DigitalOcean media offload with bidirectional sync
- RAG Content Exporter - secure REST API exposing clean content to downstream AI systems
- Social Media Generator - platform-tuned post generation across six networks
- Flowise Chatbot - on-site assistant with lead capture and AI scoring
Outcomes
- Decades of archive content made queryable through semantic and vector search
- No vendor lock-in - provider mix rebalanced for cost or quality as the LLM landscape shifts
- Editorial team reclaimed significant weekly hours across social and metadata workflows
- AI summaries published with minimal editing required
- Search relevance improved materially; zero-result searches reduced substantially
Stack
PHP 8.1+, WordPress, Elasticsearch (Cloud and self-hosted), ELSER, Cloudflare R2, custom REST APIs, multi-provider LLM integration through the AI Suite abstraction layer.