Davemaina1 learns the hard way that legal search doesn't want to live in Node

Davemaina1's fork is building search over Kenyan case law, and this stretch of work is the record of trying to run the entire AI search engine inside the app's own JavaScript runtime. It didn't go well. The embedding model - the piece that turns legal text into something a computer can match and rank - kept failing silently, crashing under load with no error to show for it. Each commit patches one symptom, and the next one surfaces.

The thread ends with the team conceding the point: the next chunk of work rips all of this out and moves the AI search into a separate Python service. So this isn't a blueprint to copy - it's a cautionary tale, useful precisely because it's an honest record of what didn't hold up.