Use cases
Find the StreamKernel scenario that already looks like your environment.
Fraud, clinical, defense, and enterprise data teams all run into the same runtime problem: model scoring, policy, evidence, and delivery need to happen before the record reaches the wrong place.
Vertical scenarios
Buyer language first, runtime mechanics underneath.
Each scenario maps a familiar operational job to the same StreamKernel boundary: source ingestion, policy, ONNX or MLflow-aware enrichment, DLQ, metrics, and governed sink delivery.
Financial Services
Fraud, AML, trading, and reporting teams that need model scoring and audit evidence before events leave controlled systems.
Real-time fraud scoring
Run ONNX inference inline on every transaction event without a round trip to a model server, including air-gapped deployment for networks that cannot call out.
AML / transaction monitoring
Apply per-event policy enforcement with OPA, attach audit headers at the batch boundary, and route flagged records to DLQ paths for review.
Pre-trade risk enrichment
Enrich order flow with model scores before delivery to execution venues while provenance labels satisfy audit trail requirements.
Regulatory reporting pipelines
Use deterministic delivery and per-batch evidence logs for SOC 2, FINRA, and internal compliance environments.
Healthcare
Clinical and device data teams that need PHI-sensitive inference, model governance, and governed routing inside their own runtime boundary.
Lab result AI classification
Classify HL7/FHIR event streams with in-JVM ONNX inference without sending PHI to an external inference endpoint.
Clinical alert enrichment
Score patient telemetry against anomaly models in the event path before routing enriched alerts to EMR or alerting sinks.
Medical device data pipelines
Govern ingestion from device event sources with model versioning, MLflow-backed promotion, and rollback when clinical models change.
Defense & Government
Disconnected and classified environments where telemetry, sensor, and ISR streams need local inference, policy denial paths, and record-level provenance.
Drone / UAV telemetry enrichment
Run real-time inference on sensor telemetry in air-gapped edge environments with single-JAR deployment and no cloud dependency.
Submarine / undersea sensor pipelines
Operate fully disconnected with a JVM-local model pool and auditable provenance attached to every record path.
Satellite downlink processing
Handle burst ingestion with backpressure management, ONNX scoring, and multi-sink delivery to classified and unclassified destinations.
SIGINT / ISR stream processing
Enforce policy before delivery, using OPA-based deny paths to keep sensitive records off the wrong sinks.
Commercial / Enterprise
Platform and data engineering teams modernizing event movement, pre-ingest AI, live feature generation, and multi-destination fan-out.
Pulsar -> Kafka migration
Swap sources without rewriting transform or sink logic, keeping the same pipeline config while the transport changes.
Pre-ingest AI enrichment
Score and label records before they land in the lakehouse, with MongoDB Vector, Delta Lake, and Snowflake delivery in one runtime.
Agent tool audit
Capture agent tool calls as governed events with policy decisions, provenance labels, DLQ paths, and audit-ready delivery.
Real-time feature generation
Compute features on live event streams and deliver them to a feature store alongside the raw record.
Multi-destination fan-out
Run one pipeline with consistent DLQ, retry, and metrics behavior across Kafka, MongoDB, Postgres, Delta Lake, and Snowflake.
Buyer triggers
When StreamKernel should enter the conversation.
The strongest fit is a team trying to score, govern, and deliver sensitive event streams without adding another fragile chain of services.
- Fraud, AML, pre-trade, or reporting flows need model scores and policy evidence before records leave controlled finance systems.
- Clinical, HL7/FHIR, or device telemetry streams need PHI-safe inference without routing records to external model endpoints.
- Defense and government telemetry must keep inference, policy, and provenance inside disconnected or classified environments.
- Enterprise migration or lakehouse teams need one fan-out path with consistent DLQ, retry, metrics, and sink behavior.
- Agentic AI programs need tool calls captured as governed events before actions affect enterprise systems.
Commercial path
Bring the scenario closest to your environment.
A short architecture review can map your fraud, clinical, defense, or enterprise data path to StreamKernel's runtime boundary.