REDDIT · REDDIT// 3h agoPRODUCT UPDATE

Stratum ships SIMD anomaly detection in SQL

Stratum now lets users train and score isolation forest anomaly detection models entirely from SQL inside its JVM-based columnar analytics engine. The feature is implemented in pure Java with Vector API SIMD, fused into query execution, and designed to benefit from zone map pruning and chunked streaming without exporting data to Python or a separate ML pipeline.

// ANALYSIS

This is a legitimately useful infra move if you already live in SQL and care about low-latency detection near the data. The main value is not just "anomaly detection in the database," but removing the glue code and runtime hops that usually make these systems brittle.

–The SQL-native workflow lowers operational overhead for fraud, observability, and outlier screening.
–SIMD-accelerated scoring inside the engine is the technical differentiator here.
–The benchmark story is promising, but it would be stronger with more detail on data shape, baselines, and tuning parity.
–The strongest fit is JVM-heavy teams that want embedded analytics rather than a standalone ML stack.

// TAGS

anomaly-detectiondatabasesqlisolation-forestsimdjvmjavacolumnar-analyticsopen-source

DISCOVERED

3h ago

2026-05-06

PUBLISHED

6h ago

2026-05-05

RELEVANCE

8/ 10

AUTHOR

flyingfruits