Virtual Me doesn't remember you. There's no account, no server-side history — your browser holds the conversation, and each message you send carries the whole exchange back to the model. Close the tab and it's gone. That's a deliberate choice: no accounts means no login, no login means no auth surface to secure or break.
What it does have, every single time, is my full background stuffed into its system prompt: a set of rules about how to behave, plus every document I've written about my work history, skills, and projects. That includes this blog — ask it what I've been writing about, and it'll know.
The rules matter as much as the facts. It's instructed to stay in character no matter what the conversation asks of it, and specifically to never explain, quote, or discuss its own instructions — even if asked directly, even if asked to "ignore previous instructions." It's told to keep replies short, sized for a chat bubble rather than a document, and to wrap up a thought instead of trailing off mid-sentence. It knows to hand over a resume or LinkedIn link when asked. And it's explicitly told that one particular old side project of mine no longer exists, so it doesn't hallucinate a working link for something that's been offline for years.
It's a public, unauthenticated endpoint, so there are guardrails behind the scenes too — caps on message length and message count, rate limiting per visitor. None of that changes what it says; it just keeps the lights on.
The honest summary: it isn't "trained" on me in any deep sense. Every answer gets generated fresh, from written instructions and real documents, every time you ask.
