AI that survives contact with production.
I help companies decide what to build with AI, then build it properly — the same way I built generative AI platforms inside a Nordic bank and shipped agents at a 50-million-user health product.
Most AI projects don’t fail technically. They fail at the decision.
Teams buy the wrong tool, prototype the wrong workflow, or build an agent nobody can evaluate. The demo works, the rollout doesn’t, and six months of engineering quietly disappears.
I’ve been on both sides of that: designing the platform an entire bank’s developers consume AI through, and sitting in procurement conversations deciding whether a vendor is worth it. What separates the two outcomes is unglamorous — clear scoping, honest evaluation, and infrastructure that a compliance team will actually sign off on.
That’s the work I do: a senior engineer who can tell you what not to build, and then build the rest.
AI strategy & advisory
Which use cases are worth funding, build vs. buy, vendor selection, and a sequenced roadmap your engineers agree with.
Agents & LLM products
RAG systems, assistants, MCP integrations and automations — taken from proof of concept to something on-call can support.
Evaluation & LLM ops
LLM-as-judge frameworks, regression suites and pipelines so you can prove a model change made things better.
Team enablement
Hands-on training and agentic-coding practice for engineering teams, from someone who writes the curriculum and the code.
Send me the problem, not a brief.
A paragraph about what you’re trying to do is enough. I’ll reply with an honest read on whether it’s worth doing and whether I’m the right person.