AI has taken the world by storm, enabling new applications, exacerbating GPU shortages, and raising concerns about the accuracy of model outputs. In this talk, I will present several projects I have worked on to address these challenges: Apache Spark, a scalable big-data processing framework; Ray, a distributed framework for scaling AI workloads; and vLLM and SGLang, two high-throughput inference engines for large language models. I will discuss how these efforts evolved from research prototypes into widely adopted industry tools, and I will conclude with key lessons learned and directions for future research.