I add AI to products that already exist, and build the automation around it that makes it actually useful.

In practice that means LLM features inside your app, document processing pipelines, retrieval over your own data, and multi-agent workflows that do real work rather than demo well. Claude API, OpenAI, and LangChain, wired into Python, Node, or TypeScript services.

The engineering discipline matters more here than the model choice: evaluation, cost control, fallbacks when the model is wrong, and a clear boundary between what the AI decides and what your system decides. I have built agent tooling and document pipelines, and I bring the same testing habits I use everywhere else.