Student wellbeing discourse on Reddit
I mined four student subreddits, classified the posts with a locally hosted Llama 3, then modeled the topics with BERTopic and bge-m3 embeddings. The context window holds 8,192 tokens, so a long post stays in one piece instead of getting chopped into fragments.