At Intuit, the ML teams encountered challenges when managing and executing inference on high-throughput streaming data. Integrating with diverse messaging systems like Kafka, Pulsar, and SQS required considerable effort, while enabling intermediate data transformations and inference added complexity. Additionally, dynamically scaling inference workflows to handle fluctuating event volumes posed unique challenges.

To address these issues, we developed Numaflow, an open-source, Kubernetes-native platform designed for scalable event processing. Numaflow simplifies connections to various event sources, empowers teams to process and infer on streaming data with ease, and integrates effortlessly with existing infrastructure. This session is ideal for ML engineers, data scientists, and anyone interested in asynchronous inference on streaming data. We’ll dive into how Numaflow eliminates bottlenecks and optimizes inference on streaming data.