Breezelearn

Problem

Model calls were slow and expensive, and every dashboard view leaned on the database to build context.

What I did

Built the Next.js platform for an AI customer-support product, and a Node.js orchestration layer connecting production APIs to Python ML services, with RAG workflows on vector databases and Redis.

What happened

250K+ inference requests a day at under 400 ms. Primary database load down 61%.