IBM’s Granite time-series models are now available in early access on Confluent Cloud, allowing businesses to run forecasting and anomaly detection where streaming data is already processed. The integration is designed to avoid copying live measurements into a separate machine-learning platform before a decision can be made.
The models run through Apache Flink on Confluent and can work across signals they have not seen during training. IBM lists forecasting, anomaly detection, similarity search, classification, gap filling and optimization among the supported jobs. Flink keeps the recent history for each series, which is essential because a reading only becomes unusual when compared with the sequence around it.
Early access begins on Confluent Cloud on AWS, with Confluent Platform support for on-premises and hybrid deployments planned next. IBM says its time-series models have been tested with partners in industries including manufacturing, food and telecommunications, but the performance figures in the announcement are company claims rather than independent comparisons. Teams will still need to validate accuracy and alert thresholds on their own equipment and business data.