Every company is being told it needs AI. Very few need a chatbot bolted onto their homepage. What usually pays off is a narrow, well-engineered feature inside a workflow your users already have: searching your documentation, drafting a reply, extracting data from a form or photo, or summarizing a long record.
I bring 23 years of production engineering to AI work. That means the same habits I used shipping apps for airline crews and insurance customers: clear requirements, monitoring, tests and cost control.
What I build
- Assistants grounded in your data. These answer questions from your docs, tickets or database (RAG), with citations.
- Document and image extraction. Invoices, forms and photos go in, and structured data comes out into your system.
- Workflow automation and agents that handle repetitive multi-step back-office tasks, with a human in the loop where it matters.
- AI inside mobile apps, using cloud LLMs or on-device models with Core ML and ML Kit for offline or private use cases.
How I keep it production-grade
- Evaluations. Each feature gets a test set, so prompt or model changes are measured, not guessed.
- Cost and latency budgets. I pick the right model size, use caching and stream responses.
- Privacy by design. I minimize the data sent, choose providers carefully and use on-device processing where appropriate.
- Graceful failure. The feature still works when the model is wrong or unavailable.
Frequently asked questions
Is my data safe if we use an LLM?
I design for that from day one. I choose providers and settings that do not train on your data, keep sensitive fields out of prompts where possible, and use on-device models when privacy requires it.
We are not sure where AI fits in our product. Can you help?
Yes. An AI opportunity workshop is a short, fixed-price engagement where we look at your workflows and data, and I recommend the one or two features most likely to pay for themselves. Sometimes the answer is that you don't need AI for it.