How I built my second brain with Obsidian and Claude (and why I needed to)

I built Loomi to solve a problem I have.

Not the vague version of that sentence. The actual problem: I consume constantly — YouTube, TikTok, LinkedIn, courses, articles — and none of it sticks. I save things and never find them again. I take notes and never look at them. The knowledge disappears into whatever app I saved it in.

So I built an app to fix it for other people. And then spent months not fixing it for myself.

One Sunday I decided to just do it.

The setup

Two tools. That's it.

Obsidian is just a nice interface for a folder on your computer. Every file in it is a normal markdown file — no lock-in, no subscription, no proprietary format. It looks complicated from the outside. It's not.

Claude does the thinking.

Here's the structure I built:

One folder called Raw Sources. I dump everything here — articles, book notes, videos, transcripts, ideas. I don't organize it. That stopped being my job.

One folder called Wiki. I never write in it. Claude does. When I drop something into Raw Sources and tell Claude to ingest it, it reads the source, figures out what matters, and builds organized wiki pages — cross-referenced, tagged, linked to everything related.

It's a system where I do the consuming and Claude does the organizing. So far it's working better than anything I've tried.

The moment it got real

I wrote a file called memory-seed.md. Just me, talking honestly about where I am right now — what I'm building, what I'm avoiding, what I'm afraid of, what I want. Stream of consciousness. No structure, no editing.

I told Claude to ingest it.

It came back with wiki pages it had built from what I wrote. One of them was called avoidance-as-fear-not-bandwidth.

It had named a pattern in my own words before I'd consciously named it — that when I avoid things, it's not because I don't have time. It's because I'm scared. And it had built a whole page around it, with a table of everything I said I was avoiding and what I said the reason was.

I didn't ask it to do that. It just saw it in what I wrote.

Then I opened the graph view — a visual map of every page and how they connect. In the middle: a node called mariana. Around it: every project I'm building, every client I work with, every pattern Claude had named, every strategy it had pulled from my words.

One file in. One afternoon.

What I was worried about that turned out to be nothing

That it would be too technical. It wasn't. I got one permissions error setting up Claude Code. Claude told me how to fix it in one message. That was the hardest part.

That I'd need to build the whole thing before it was useful. You don't. You start with one file — honestly just a brain dump about yourself — and it grows from there.

That I'd have to maintain it constantly. The only rule is: drop sources in, ask Claude to ingest. That's it. The system does the rest.

Six months from now, when I want to know everything I've learned about building a SaaS product, I won't search through folders. I'll just ask. And the answer will come from my knowledge — the stuff I've actually read and lived through.

That's what I wanted when I built Loomi. Turns out I could build it for myself in an afternoon.

Mariana