Before I ever wrote code, I wrote poetry. It’s always been my way of sorting out the abstract — feelings, memory, mood, the in-between. I’ve even self-published a book on Amazon. So yeah, poetry isn’t just a hobby to me — it’s part of who I am.
That’s what made me wonder: what happens when someone like me teaches a machine to write poetry? Would it feel mechanical, or could it actually create something human? That’s howPoetAI happened — a project where I trained a deep learning model to generate original poems.
The setup used stacked LSTM layers, trained on real poetry datasets, with embeddings and dense layers to capture structure and tone. It trained over 170 epochs, hit 95% accuracy, and produced output that was… well, weird and wonderful.
One of the poems it generated on the theme of sadness included this:
“brings me chocolate from the pentagon / dark chocolates shaped like…”
and then it just spirals. But in a way that feels kind of real.
It wasn’t perfect. But it surprised me how often it landed on something close to emotional truth. The chaos made it feel honest — like the model was stumbling into beauty without knowing it, which honestly, isn’t that far off from how human poetry works too.
PoetAI didn’t replace my voice. But it gave me a strange new one to play with. As someone who writes code and poems, I liked that it blurred the line between the two. Sometimes I guide the words. Sometimes I let the model go wild.
Poetry isn’t about rules. And neither is building cool stuff with AI. This project reminded me how those two ideas actually speak the same language — just in different forms.