AI-enabled tools defined and coordinated across strategy, design, research, QA and development: an AI content-tagging agent, AI-assisted translation inside the CMS, AI-search (GEO) visibility, a voice concierge in definition and Notion AI agents.
Client brand team, SEO, development and AI engineers
Period
2025 – present
Context
Applying AI where it saves time in content operations and findability, while keeping a human in control of the output.
Outcome
4 AI initiatives delivered, 1 in definition
≈324 h of manual tagging reduced to seconds per run
Human validation on every AI output
From a 45-market programme to a community platform
Use of AI
Defined and coordinated AI-enabled tools: content tagging, translation, AI-search visibility and a voice concierge.
Where each piece stands
Delivered
AI tagging tool
I defined the agent and coordinated strategy, design, research, QA and development. It tags a content library (articles, recipes, experts, chefs) against the brand’s taxonomies with ~80% precision and a final human validation. Estimated manual effort ≈324 h, calculated from the time a content manager takes per item × total content (2,150 articles, 650 recipes, 100 experts, 150 chefs); the agent runs in seconds.
Estimated manual effort versus the agent, and the human-in-the-loop flow
Delivered
AI-assisted translation inside the CMS
An AI translation tool integrated into the Drupal CMS, so editors translate content for new markets without leaving the platform. A prompt set in the back end drives the translation, and the editor chooses the tone when translating. Every translation still goes through manual validation before publishing, so speed never replaces quality control. Result: shorter translation time for each market rollout.
AI-assisted translation inside the CMS, always validated by a person
Delivered
GEO / LLM visibility
Generative search engines (ChatGPT, Gemini, Perplexity, Claude) answer questions by citing sources, so a site has to be readable and citable by them. On the 45-market programme I coordinated the platform-level work: schema markup, HTML landmarks, breadcrumb schema and an llms.txt file. On the community platform I coordinated a diagnostic of AI readability: llms.txt, JSON-LD, robots.txt rules for AI bots, server-side rendering, FAQ pages, content chunking and WebMCP readiness. Visibility is measured with prompt-based checks across the main AI assistants.
In definition
Voice-activated AI concierge
Product definition and roadmap of a voice-activated assistant for community members. It lets them find restaurants, listen to recipes while cooking and get guided through articles, using only the brand’s approved content through the CMS API (user → agent → CMS API). Planned in three phases: restaurants, recipes and articles first, then governed personalisation, then an access and benefits layer. Currently in definition (scoping, architecture and roadmap); not built yet. My role: project definition, technical alignment, data review, timeline, strategy and client contact.
Voice-activated concierge: controlled data flow and a three-horizon roadmap (in definition)
Delivered
Notion AI agents
Three Notion agents built for a client to cut manual admin: logging email messages into Notion, drafting replies and automating follow-ups. Estimated time saving: up to ~10 hours of manual work per week.
Skills shown
AI product definition
Cross-functional coordination
Human-in-the-loop design
GEO / AI search visibility
What I learned
At 80% precision, human validation is part of the product; the time saved is net of review.
AI is deployable for a brand when it is constrained to approved content.
Measure the manual baseline first: the tagging case was provable because we had the 324 hours.
For a new discipline like GEO, start with a diagnostic and measurement before investing.