Building an AI presentation tool that handles 500+ daily generations required solving real-time streaming at every layer. Instead of users waiting 60 seconds for a spinner, SlideMaker streams each slide as it’s generated using SSE, Redis pub/sub, and React. This is a complete architecture breakdown: why I chose Flask over FastAPI, how Redis handles four different jobs, and the patterns that let one EC2 instance serve hundreds of users daily.
