Blueprint 1 · Real-Time ML Inference

ML inference at the data layer. No external model serving required.

Execute ML inference directly within streaming data pipelines—where transactions happen, before data moves.

Back to all blueprints »

Try it now
<100ms

detection latency, Kafka path

4

inference paths validated on PoC infrastructure

100%

data on-premises; no cloud ML service dependencies

ARCHITECTURE

How it works

Blueprint 01

How data moves

Architecture flow

01  SOURCE  

Transaction events write directly to EDB Postgres® AI (EDB PG AI) as the operational system of record. EDB Postgres Distributed is the source of truth—four downstream analytics and inference paths read from it via WAL, eliminating the need to move data before processing begins.

  • Debezium
02  STREAM  

Debezium captures WAL changes and streams them to Kafka. Kafka routes events to three parallel downstream paths: Kafka direct, ClickHouse (via CDC), and RisingWave (via CDC). A fourth path—PGAA—reads the WAL independently of Kafka. (Note: Redpanda is a validated Kafka-compatible alternative.)

  • Kafka/Redpanda
03  AGGREGATE  

RisingWave computes streaming materialized views: rolling spending patterns, geo-anomaly scores, and velocity checks. ClickHouse runs 90-day behavioral baseline comparisons for complex historical aggregations.

  • RisingWave
04  INFER  

XGBoost ML models execute inference across all four paths with measured latencies from the PoC: Kafka direct <100ms TTDF;  PGAA ~1500ms (WAL → Iceberg); ClickHouse ~3000ms; RisingWave ~4200ms. (Notes: KServe + NVIDIA NIM is supported in EDB PG AI but not in the current open source release. MLflow is supported but not integrated in the current release.)

  • NVIDIA NIM
05 GOVERN  

Human-in-the-loop decision gates enforce policy on high-risk agent actions before execution. Lakekeeper (Vakamo) provides the Iceberg REST catalog—governing data lineage and access control across all four inference paths.

  • LAKEKEEPER (VAKAMO)
05  PERSIST  

Fraud predictions and operational alerts write back to EDB PG AI. Historical inference results sink to Iceberg tables on MinIO for audit, model retraining, and longitudinal analytics.

  • MINIO

Blueprint 1 · PARTNER STACK

Validated partners in this blueprint

Airflow (Astronomer)

Orchestrates the full lakehouse pipeline — ingestion, dbt transformation triggers, and WarehousePG analytical load — reliably at scale.
Transformation BP 01 BP 03
View integration

dbt

Transforms source data natively in Postgres, aligned to BCBS 239, EBA ITS, OMOP CDM, with full lineage per run.
Transformation BP 01 BP 03
View integration

Grafana

Monitors Airflow pipeline health, WarehousePG query performance, and lakehouse telemetry in a unified real-time view.
Visualization BP 01 BP 02 BP 03
View integration

Jupyter

Connects data scientists directly to governed WarehousePG and Iceberg data for population health and regulatory research.
Development BP 01 BP 03
View integration

Kafka / Redpanda

Streams high-velocity event data directly into EDB PG AI pipelines as queryable, transactional records — the ingest backbone for all three inference paths.
Streaming BP 01 BP 02
View integration

KServe + NVIDIA NIM

Deploys GPU-optimized model endpoints that execute inference directly against live Postgres data — co-located, not external.
AI serving BP 01
View integration

Lakekeeper (Vakamo)

Governs Iceberg table metadata across the lakehouse — unified access control, lineage, and discoverability for regulatory consumers.
Storage BP 01 BP 03
View integration

Langflow

Orchestrates LLM reasoning pipelines over Analytics Engine retrieval results, using EDB PG AI as the agent runtime and state layer.
AI/ML BP 01
View integration

MinIO

Provides sovereign S3-compatible object storage for Iceberg-formatted inference results, model artifacts, and long-term analytics data.
Storage BP 01 BP 02 BP 03
View integration

MLflow

Tracks every model version, experiment, and deployment decision tied to EDB PG AI pipeline runs — full lineage from data to inference.
AI/ML BP 01
View integration

RisingWave

Computes streaming materialized views in SQL, producing aggregated features for the ML inference layer in real time.
Streaming BP 01
View integration

INDUSTRY USE CASES

Blueprint 1 in production

  • BFSI
    BFSI

    Real-time fraud detection

    A tier-1 bank processes 50,000 transactions per second. Batch-based fraud detection carries a 15-minute detection window. Deploying XGBoost within the Kafka streaming path cuts detection latency from 15 minutes to 500ms. Three parallel inference paths provide immediate blocking decisions and behavioral context enrichment simultaneously. All data remains on-premises—no cloud ML service dependencies.

Validated deployment environments

Runs on-premises, on IBM Power, on EDB engineered systems, or on any cloud—with consistent Postgres interface across all environments.

Logos

Try it now—Blueprint 1 is open source.

The full BFSI fraud detection implementation is on GitHub.
Deploy it, fork it, or build on it.

TRY IT NOW
Talk to a solutions engineer