The terminal on this site answers questions now. Type anything that isn't a command and an automated recruiter replies, using only my résumé. It's there for interviewers.

It runs on bnm-chat, the chatbot framework I built for Brooks New Media client sites, the same one behind the chat on brooksnewmedia.com. Before it went live it had to pass a 110-question test three runs in a row: questions it must answer from the résumé, plus ones it must decline, including prompt injection and attempts to pull out its instructions.

The first model was a small one running locally. It passed the test, and then I read the answers. Asked what I did at Google, it said it didn't know, which accepts that I worked there. It also pasted example sentences from its own rules into replies. I moved it to Claude Haiku 4.5 and changed the test so both mistakes fail.

It declines off-topic questions and won't make up a salary, references or personal details. Ask which model it is and how it works, and it will tell you.

Try asking it something my résumé doesn't say.