PoetAi is app i started building to leverage the rise in Large Language Models, particularly in the realm of poetry which is an area im deeply passionate about. I also was an opportunity to refresh and explore some interesting technologies which i don’t normally have the chance to play with as much. The project is still in development with a basic alpha in testing focused on just using AI to analyse existing popular classic poems.
Technology stack
I took a very basic system design, with mobile android client hitting a basic service which pull all relevant poetry date from a postgres DB. Further one of my core interest was to also explore the new jetpack compose framework from google. I also leverage NodeJS/Express and most joyfully typescript, given the last time i built a backend solution in JS, the move to a more strongly typed language was definitely appreciated, i felt more certain about what i was implementing… Since i was working on my AWS certification i decided to host it all on AWS, just to get refresher on their offering.
Android, Kotlin, Jetpack Compose, Postgres, NodeJS, Typescript, AWS
Result
What was surprising about the result is not so much that it made sense, but that i provided a good breakdown of the basic historic context of the poems, given the hallucination problem with LLM’s, i was definitely interested to seeing how much it would be off especially given the large variance in how one can interpret poetry. What better way to test this but to analyze its performance on the classics, to the right is an extract from PoetAi analysis of the famous “Death Be Not Proud” by John Donne. As one can assess not only is it fairly accurate it provides the most basic layer of understanding for any general reader to work from.
