Hello, We're evaluating Apache Pinot for a fraud detection use case and wanted to get the community’s input on whether Pinot is a good fit, or if we’d need to pair it with something like Flink for stateful or more complex aggregations. Here’s what we’re trying to support: Aggregates • COUNT(), DISTINCT COUNT(), AVG(), MAX(), MIN(), SUM() • Statistical (nice to have not mandatory):

STDDEV()

,

Z-SCORE()

(for anomaly detection) • Time-based:

TIME_DIFF()

,

FIRST_SEEN()

,

LAST_SEEN()

• Geo:

DISTANCE()

between two coordinates,

TIMEZONE_DIFF()

Requirements • Support windowing for up to 12 months • Ingesting 10K events/sec • Query latency target: <200ms We know Pinot handles real-time ingestion well with fast OLAP queries, but for some of the stateful or time-windowed logic (e.g.,

Z-SCORE()

,

TIME_DIFF()

,

FIRST_SEEN()

), would it require using Flink or another stream processor, or can this be handled natively in Pinot (via UDFs, rollups, pre-aggregated tables, etc.) while still meeting requirements? Would love to hear any thoughts, lessons learned, or best practices from folks who’ve done similar Thanks in advance

Thanks for sharing the details. Pinot does support window functions natively since 1.0. However, given your read requirements you may want to precompute them.

Curious if Flink was chosen for this use-case or another stream processor 🙇‍♀️