Vector Similarity, Zero Client JS: Decoupled Analytics on a Side Project Budget

The author built a leaderboard for DumbQuestion.ai that groups semantically identical questions using vector embeddings and Qdrant's cosine similarity, rather than simple string matching. To keep costs minimal, the vector processing runs asynchronously via a sidecar container and GCP Pub/Sub on a home Mac Mini server, with results cached as static JSON in Cloudflare R2 for essentially free page loads at any scale.