The Work
At Lenovo I pioneered Generative Engine Optimization methods to optimize the brand's visibility in ChatGPT, Perplexity, and other AI search platforms, and established the monitoring systems that track Lenovo brand mentions and product recommendations inside AI-generated responses. You can't optimize what you can't measure, so measurement came first.
That work crystallized into Conversation Optimization, my own GEO strategy for earning presence in AI conversations, not just AI search results.
Most GEO practitioners work from the outside, guessing at how the engines behave. I work from both sides. I fine-tuned my own GEO-specialist model (a Qwen3-30B trained on 5,000+ custom examples of Conversation Optimization) to bake the strategy directly into model weights. And I built 4ort.xyz, a Wikidata-grounded knowledge graph of ~40M entities, which is exactly the kind of infrastructure answer engines use to decide what's true and who to cite. When you've built an entity graph yourself, citation optimization stops being folklore.
The through-line: GEO is the successor discipline to SEO, and the practitioners who win it will be the ones who understand the AI systems as builders, not just observers.
Capabilities
- Conversation Optimization: my own GEO strategy for visibility in ChatGPT, Perplexity, and Claude conversations
- Citation optimization: structuring content and authority signals so AI systems cite you
- Structured data for AI: entity layers and machine-readable formats designed for LLM consumption
- AI response monitoring: systems that track brand mentions and product recommendations across AI platforms
- AI search ranking factors: hands-on research into what generative engines actually reward
- Model-side understanding: LLM fine-tuning and knowledge-graph construction: GEO informed by building the machinery itself
Proof of Work
→ Qwen3-30B GEO fine-tune on Hugging Face: Conversation Optimization baked into model weights.
→ 4ort.xyz knowledge graph: ~40M entities, grounding AI agents in verified facts.