AI read my 150 pages of journaling

Hello Reader,

I recently experimented with vision language models to analyze 5 years of my handwritten journal entries. Trust me, you're going to love this one 😉.

Every 5 years or so, I feel the urge to reassess my life - figure out where I'm going and who I'm becoming.

I've been journaling on a regular basis. Just with pen and paper, writing down what I'm thinking and what's happening.

Then I had an idea.

What if I scanned all those pages, used a vision model to extract the text, and then asked an LLM questions about myself? With actual patterns across hundreds of pages, I hoped to shed some light on who I am and how I think.

Being a data nerd, this whole stopped-guessing-and-let-the-data-speak vibe is something I couldn't resist..

Here's how I did it:

First, I needed to get the text out of those handwritten pages.

I scanned ~150 of them with my phone and ran each image through a local vision language model via Ollama. I asked it to transcribe every word exactly as written, describe any sketches or drawings, save output to a structured JSON file.

Quick note: everything runs locally, so my secret diary stays on my laptop. I'm not ready to hand my private thoughts to big tech just yet. 🙃

After one model (llava4) nearly killed my laptop and another confidently hallucinated things I definitely never wrote, I landed on Qwen3VL. Stable, accurate, and completely unbothered by my deteriorating handwriting.

It does occasionally struggle with my fancy calligraphy 😁. But honestly, it's pretty impressive!

Some extracted words are not accurate, especially when I do fancy calligraphy.

Once I had a massive blob of journal text, I passed it into Llama 3.2 and started asking questions:

What patterns do you see in what motivates me?
What do I seem to struggle with repeatedly?
What could be my ikigai?

The outputs were surprisingly good. It found patterns like:

I'm motivated by personal growth, creativity, financial independence, and helping others.

I struggle with overthinking, self-doubt, and work-life balance.

And my ikigai (i.e. purpose at the intersection of what you love, do well, and can get paid for) apparently points to teacher, content creator, or artist.

Which is... kind of... exactly what I'm already doing 🧐.

This was a fun excuse to experiment with local, open-source AI models. I'm amazed how capable they've become.

Again, LLMs aren't perfect, but they're surprisingly good at finding patterns. And what better patterns to find than insights into your own life and the way you think?

If you want to learn more about this project, check out the full walkthrough in my newest video.