Sidenote

This bot is just a parody of several governments' social credit scoring systems which some may describe as Orwellian.

Three Versions

My first version of this bot was crude: extract specific words and phrases using compromise. This involved a lot of manual definitions and wasn't flexible. I rewrote it around NLP.js to dip my toes into machine learning. With an iota of training data compared to models like GPT and Llama, my bot was pretty unreliable - tons of false positives and unexpected pattern biases. It was a great learning experience, though. It made sense to approach the model-training feature with Test-Driven Development and I also got more familiar with AI. I finally decided to hack together a GPT wrapper. It took less than an hour. This public version is very minimal, but you can modify the prompt to fit just about any scenario you want. After this experience, I might start building some more of these AI wrappers. They're dead simple and extremely powerful.

Favorite Feature

This is the software equivalent of a complaint jar that automatically empties itself.