Data science teams often use Apache Spark on Dataproc for large-scale data preparation. Integrating these pipelines with machine learning models, especially for inference, has been complex. To simplify this, a new open-source Python library called Dataproc ML has been developed. This library seamlessly connects Apache Spark jobs with popular ML frameworks and Vertex AI features. It follows a familiar SparkML-style builder pattern, allowing users to configure models and apply them to DataFrames using a transform function.Dataproc ML enables applying generative AI models like Gemini to Spark DataFrames for tasks such as classification and summarization at scale. It also supports running inference with PyTorch and TensorFlow models directly from Google Cloud Storage. This is achieved by loading model weights and defining pre-processors for batch inference without separate serving endpoints. The library is built for performance using optimizations like vectorized data transfer and connection re-use. Future plans include support for Spark Connect, more Vertex AI integrations, and additional performance optimizations. The library aims to simplify AI/ML inference directly within Spark environments.