How would you design batch and streaming data pipelines using Apache Beam, Spark, and Ray for Character.AI?