Hi everyone, I’m excited to share Otter-Streams, an open-source SDK and framework I’ve been building to make real-time machine learning inference first-class in Apache Flink. Otter Streams focuses on: • Streaming-native ML inference (low-latency, under evaluation and testing) • Clean abstractions for model loading, feature handling, and inference • Seamless integration with Flink DataStreams • Support for embedding inference logic directly into Flink pipelines (no external serving hop) The goal is to simplify how we run online ML inference at scale, while staying idiomatic to Flink’s execution and state model. Documentation: https://martourez21.github.io/otter-streams/ GitHub Repository: https://github.com/martourez21/otter-streams I’d really appreciate: • Feedback on the API and design • Suggestions for improvements or missing features • Ideas from folks running ML in production on Flink I’m also very open to *open-source collaboration*—issues, discussions, PRs, or design feedback are all welcome.