Turning Your Domain Expertise Into a Custom GPT

A Claude Project or custom GPT is just a pre-baked system prompt plus a folder of reference documents. Most working AI tools are 80% prompt engineering and 20% data. As a 20-year veteran, you have the data — your binders, internal manuals, sample work, training decks. Most of which the public model has never seen.

The 4-hour build

1. 1Pick a job you used to do that took an hour and was repetitive (proposal review, claim triage, lesson plan QA).
2. 2Write a 1-page system prompt: who the assistant is, what it should always check, what tone to use, what it should never do.
3. 3Upload 5-10 anonymized examples of inputs and the gold-standard output you'd produce.
4. 4Test it with 3 fresh inputs from real life. Note where it's wrong. Update the prompt.

Anonymization is non-negotiable

Showing it in interviews

In an interview, you can say: 'Want to see what I mean? I built a small assistant that does first-pass [task] using my industry experience as the guardrails. Here's a 90-second demo.' This converts an abstract claim into a concrete artifact. Most candidates can't do this. You now can.

The big idea: a domain-loaded assistant is the cheapest portfolio piece you'll ever build, and the highest-leverage thing you can show a hiring manager.