Where I help
Most AI work that ships isn’t a platform — it’s a small, focused tool that does one job well. My most concrete pattern lately is local-running utilities that automate catalogue input for catalogue-heavy merchants: take a supplier’s spec sheet, PDF catalogue, or messy CSV, and produce structured Shopify-ready records with the team’s review in the loop. The tool runs on the merchant’s machine, so the catalogue data never leaves the building.
I’ll scope, design, and build the first working slice; from there it’s usually co-design with your team.
You probably want a Plus-Fit Assessment first if any of these are true:
- You have a catalogue-input bottleneck — supplier data arriving in PDFs, spreadsheets, or scraped HTML, and a human team transcribing it into Shopify.
- You’re considering a vendor “AI platform” and you’re not sure whether you actually need it.
- You want retrieval over private data (Q&A against your docs, code, or product data) and the brief skips over evals, observability, and what “good” looks like.
- You want an internal copilot for a specific role (merchandising, support, ops) but the scope keeps drifting toward a generic chatbot.
What the Plus-Fit gives you for AI work
- Job framing. What’s the human doing today? What part is genuinely worth automating, and what stays human?
- Smallest-model recommendation. Latency, cost, observability — bigger and newer is not always better.
- Evals first. A small but real evaluation set that catches regressions before you ship.
- A narrowly-scoped first artefact. One bounded tool, deployed, measured, iterated — no platforms, no roadmaps.
When AI is the wrong answer
Often it is. If a problem is solvable with a query, a form, or a rule, I’ll say so before we sign.