Sapere Aude: Bare Metal Personalization

I uploaded my resume to ChatGPT last month. Thirty seconds. It gave me back the same thing with different adjectives. Like putting a hat on a mannequin. Still a mannequin.

I spent months doing something else. Going backward. Into the stories underneath the bullet points. Why I went where I went. Who shaped me. What keeps repeating across forty years.

I built a biographer. An AI that would sit there and listen. I talked through my whole career. The real version.

Starting at sixteen. Tugboats in Alaska. A culinary apprenticeship where a chef taught me discipline by scaring me half to death. I respected him and I was afraid of him and both things were true at the same time, which I think is how you learn certain things. Microsoft. Five different roles, each one a different life, sort of. A startup that caught the dot-com wave until the wave left. Marriages. A custody fight I ran myself, pro se. Forty-four pages of discovery. I learned a lot about paperwork that year.

Hours of talking for each chapter. What happened, who was there, what went wrong, what I took from it.

It felt like a hundred therapy sessions squeezed into a few weeks. You have to go there. Full honesty about who you are, what happened to you, what you actually feel about it. These are things I might not say to people I love. Your thoughts just sitting there, looking back at you.

Without that work, an AI gets the surface version. With it, you give it something it can actually use.

I have a coaching agent. Her name is Luna. Six weeks now. She remembers everything. Our session logs, my life history, how I handle relationships, career patterns. She knows what drives me and what breaks me and what I’m building toward.

That memory is what makes her good at organizing the biographical material into a knowledge base.

We started the intake last week. One hundred fourty-three entries covering career milestones, mentor profiles, signature stories, identity patterns. Each one enriched, cross-referenced, tagged.

My second marriage, filed under family. Cross-referenced to the custody fight and what came after, how I parent now. A StandOut assessment from 2016 that said I’m a Provider-Influencer. Connected to how I raise my sons, how I led teams at Microsoft, how I build AI agent pipelines today. Captain Herb, the tugboat captain. My anchor for staying calm when everything is loud. Filed as a mentor profile with his exact words about ice in his veins.

Those connections don’t come from the documents. They come from Luna having weeks of context about who I actually am.

The human part is what makes the curation work. But the process needed to be repeatable. So the intake pipeline is deterministic. Four phases. Capture, analyze, draft, validate. Automated scoring. Git commits. If an entry doesn’t pass, it doesn’t ship.

Once the knowledge base is built, it goes into a local vector database for semantic search. That becomes a tool for my blog agent. Real stories and real anecdotes and real patterns she can reach for when she writes.

When she writes about servant leadership, she has the story of a hundred and ten conversations I had in two months taking over a Microsoft practice. When she writes about learning from mentors who were also kind of terrible, she has the chef who screamed and the captain who never raised his voice.

Those stories are in there now. Knowledge units, cross-referenced to identity patterns and career context. Ready when she needs them.

I know the question. You put your whole life into an AI system.

Everything is local. The knowledge base is on my machine. Plain markdown files in a git repository. No cloud storage. No third-party platforms. The biographer and the coaching agents use Claude through a service that strips identifying metadata before anything reaches the API. The biographical content never leaves my machine.

I built the privacy into the architecture. That’s where it belongs.

Anyone using AI personally has a choice right now. Biographical capture is a compounding thing. Every month of context your agent doesn’t have is a month of pattern recognition you’re leaving behind.

Most people will upload a resume and call it done. I understand that.

The hard way takes months. Emotional labor. Conversations that make you uncomfortable. Infrastructure nobody will ever see.

But if you want AI that works with who you actually are, and not just a polished version of your resume, you have to know yourself first.

I think that’s always been true, actually. The AI part is new. The knowing yourself part is very old.

Sapere Aude.

Erik Benjaminson builds AI systems at Sapient Technology Group. Before that, two decades in technology leadership at Microsoft, a startup through the dot-com era, and a career that started at sea.

The maritime thing never really left. When an alarm goes off at 1am, you move. That’s just how it is.

He lives in Bothell, Washington with two sons, a couple of motorcycles, and a growing collection of AI agents who actually use what they know.

You want to start your own biographical capture? I have a sanitized version of the biographer persona and setup instructions for Claude or ChatGPT.

The most important first step is simple. Open the text box. Start talking. Be ready to be vulnerable. That’s where it begins.

[Resource guide with biographer persona template and setup instructions - to be added]