The best customer to build for is yourself.

Ever since I started working with LLMs, the first thing I wanted was an AI personal assistant.

Let’s be honest, Jarvis from Iron Man planted that seed in all of us years ago.

I’ve seen a ton of projects chase that dream in their own ways.

Some go OCR-heavy for people glued to notebooks.

Others build CV workbenches to help with hardware dev.

And then there’s Poke; more of a digital companion that lives in your text threads (Poke VERY cool).

All solid ideas.

But none really clicked for me.

So I did what most builders eventually do; I made my own.

A dead-simple version focused on one thing: being a second brain that remembers things for me.

A few hours later, with some help from Cursor, ChatGPT, and Ollama, I had a solid MVP running.

(Pro tip: running the same prompt across multiple LLMs is an underrated experiment. The variance in answers teaches you a lot about model behavior.)

My biggest learning?

Writing PRDs for LLMs completely changed the output quality.

When you design your instructions like you’d write for a human engineer, the models actually deliver on the first try.

Below are some screenshots of the app (in season-appropriate Halloween colors 🎃).

Simple, but it already feels like a glimpse of the future I wanted to use every day.