How does RisingWave support real-time feature engineering?

RisingWave computes features from streaming data using SQL — time-windowed aggregations, joins across multiple streams, and statistical functions. Features are maintained in materialized views and served with sub-second freshness to ML model serving systems.

Can RisingWave feed features to online ML models?

Yes. RisingWave's PostgreSQL compatibility means any ML serving system with a PostgreSQL driver can query feature values directly. Features update incrementally as new data arrives, ensuring models always use the freshest signals.

Use Case

Process Millions of Sensor Readings with Simple SQL

Ingest, aggregate, and analyze IoT telemetry at scale — from factory sensors to connected vehicles to smart infrastructure — using PostgreSQL-compatible streaming SQL.

Live Telemetry Hub

Streaming

Connected Devices

48,291

all reporting

Readings/sec

2.4M

sensor data points

Anomalies (1h)

flagged for review

Processing Lag

120ms

near real-time

See Live Demo

Trusted by 1,000+ Data-Driven Organizations

for Real-time Analytics

Trusted by 1,000+ Data-Driven Organizations for Real-time Analytics

The Problem

Is Your Telemetry Data Sitting in a Queue While Machines Drift Out of Spec?

IoT platforms collect data fast, but analyzing it is another story. Most teams dump telemetry into a data warehouse and run batch queries — by which time a machine has already overheated or a vehicle has already broken down.

With RisingWave

Analyze Every Reading As It Arrives, Not Hours Later

RisingWave processes millions of sensor readings per second using familiar SQL. Create materialized views that continuously compute rolling averages, detect threshold violations, and trigger alerts — all with sub-second latency.

High-Throughput Ingestion

Ingest millions of events per second from Kafka, MQTT, or any streaming source. No data loss, exactly-once semantics.

Windowed Aggregations

Compute tumbling, sliding, and session windows over sensor data. Moving averages, percentiles, and anomaly scores — all in SQL.

Real-Time Anomaly Detection

Define anomaly rules that evaluate continuously. Detect vibration spikes, temperature drift, and pressure excursions within milliseconds.

See RisingWave in Action: IoT & Telemetry

See how RisingWave processes real data in real time — not a recording, not a simulation.

A precision aerospace parts manufacturer runs 120 5-axis CNC machines producing turbine blades. Spindle vibration signatures predict bearing failure 4-6 hours before catastrophic breakdown — but only if analyzed continuously, not in hourly batch windows.

⚠A spindle bearing on CNC-078 failed mid-cut, scrapping a $45K titanium turbine disk and causing 14 hours of unplanned downtime.

LIVEspindle_vibration

machine_id	spindle_rpm	vibration_rms	dominant_freq_hz	bearing_temp_c	coolant_flow_lpm	ts
CNC-078	12000	2.1	1200	42.3	18.5	2024-03-15T08:00:01.000Z
CNC-078	12000	2.8	1140	44.1	18.4	2024-03-15T08:05:01.000Z
CNC-078	12000	3.5	1020	47.6	18.2	2024-03-15T08:10:01.000Z
CNC-078	12000	4.2	940	51.8	17.9	2024-03-15T08:15:01.000Z
CNC-078	12000	4.8	890	55.2	17.6	2024-03-15T08:20:01.000Z
CNC-031	15000	1.4	1500	38.7	20.1	2024-03-15T08:05:01.000Z

Streaming SQLRunning

▶Ingest spindle vibration telemetry from Kafka✓

CREATE SOURCE spindle_vibration WITH (
  connector = 'kafka',
  topic = 'cnc.spindle.vibration',
  properties.bootstrap.server = 'broker:9092'
) FORMAT PLAIN ENCODE JSON;

▶Match vibration patterns against failure signatures✓

bearing_healthauto-updating

machine_id	vibration_trend	freq_shift	health_score	predicted_failure	action
CNC-078	RISING	310	15	IMMINENT	DISPATCH_MAINTENANCE
CNC-031	STABLE	0	97	NULL	MONITOR
CNC-112	STABLE	0	97	NULL	MONITOR

✓RisingWave detects CNC-078's vibration RMS rising from 2.1 to 4.8 mm/s with dominant frequency shifting from 1,200 Hz to 890 Hz — classic inner race defect signature. Maintenance dispatched 5 hours before failure.

View all demos

Why RisingWave

Turn Raw Telemetry Into Actionable Intelligence

Use streaming SQL to ingest, aggregate, and analyze IoT data at scale — from edge devices to operational dashboards — with sub-second latency.

Prevent Equipment Failures

Detect vibration anomalies, temperature drift, and pressure excursions in real time — before they cause costly downtime or safety incidents.

Reduce Data-to-Decision Latency

Go from raw sensor reading to actionable insight in milliseconds, not hours. Make operational decisions based on what is happening now, not what happened yesterday.

Scale Without Infrastructure Complexity

Handle millions of readings per second with standard SQL — no custom stream processing frameworks, no JVM tuning, no ops burden.

Process Millions of Sensor Readings with Simple SQL

Is Your Telemetry Data Sitting in a Queue While Machines Drift Out of Spec?

Analyze Every Reading As It Arrives, Not Hours Later

Turn Raw Telemetry Into Actionable Intelligence

Ready to Build Real-Time IoT Analytics?