AI & Automation Consulting

The boring, repeatable ones: processing documents and invoices, extracting data from PDFs and emails, triaging inboxes, answering questions against your internal documents, drafting first-pass reports, and gluing together tools that don't talk to each other. If a task follows a pattern and eats hours every week, it's probably automatable. If it requires real judgment every time, it probably isn't — and I'll say so.

No. Most useful AI automation projects are one to six weeks of work built on existing models like Claude — no training, no data science team, no six-month pilot. The right first project is the smallest thing that removes a real chunk of weekly busywork, and you measure whether it worked before spending more.

It can be, if it's set up properly. I build on commercial APIs (like Anthropic and OpenAI) under terms where your data isn't used to train models, keep your data inside accounts you own, and avoid sending anything sensitive that doesn't need to leave your systems. Part of every engagement is walking you through exactly where your data goes.

Mostly the Claude and OpenAI APIs for the model layer, with n8n, Zapier, or custom scripts for the workflow plumbing — chosen per problem, not by loyalty. I'm deliberately vendor-neutral: you get whatever combination solves your workflow cheapest and most reliably.

We start with the workflow, not the technology: identify the repeatable task, prototype the smallest version, and measure it against the manual process. If the prototype doesn't clearly win, you stop early and cheap. I turn down projects where I don't think AI is the right answer — a strategy deck with no working software helps nobody.