Kafka that holds up in production, not just in a proof of concept.

A streaming platform lives or dies by the architecture decisions made up front. We think through partitioning, delivery guarantees, schema contracts and state with you, then run the platform under an SLA.

Where streaming makes the difference
security

Real-time fraud detection

Suspicious transactions are detected and stopped instantly, not at end of day.

inventory

Inventory & supply chains

Stock, shipments and demand synchronised in real time across all sites and systems.

smart_toy

Feed AI with fresh data

ML models and RAG systems get live data instead of stale snapshots, the basis for relevant AI.

sync_alt

System integration & CDC

Change data capture from databases in real time, systems stay in sync without nightly ETL runs.

Deep dive · core concepts

What matters under the hood.

A streaming platform stands or falls on the right architecture decisions. Here are the concepts we think through cleanly with you — before the first line of code.

01 — Partitioning

Ordering vs. parallelism

The partition key determines ordering guarantees and throughput. Chosen wrongly it means hot partitions or lost ordering. We design keys around your domain logic.

02 — Delivery guarantees

At-least / exactly-once

With payments every message counts exactly once. We use idempotent producers, transactional writes and consumer offset strategies matched to your consistency needs.

03 — Consumer groups

Scaling & rebalancing

Load distribution across consumer instances, clean rebalancing without double processing, lag monitoring as an early warning for overloaded pipelines.

04 — Schema evolution

Contracts that hold

Avro, Protobuf or JSON Schema with a schema registry and compatibility modes. Producers and consumers evolve without anything breaking.

05 — Stateful processing

Windows, joins & state

Aggregations over time windows, stream-stream and stream-table joins, materialised state with Kafka Streams or Apache Flink, event time instead of processing time.

06 — Retention & compaction

The log as source of truth

Time-based retention for events, log compaction for state topics. Replay from the log makes event sourcing and disaster recovery a given.

history_edu

Event sourcing

State is derived from the event log, with a complete audit trail and time travel included.

call_split

CQRS

Write and read paths scaled separately, with optimised read models fed from the event stream.

account_tree

Saga / outbox

Distributed transactions across services, consistent without two-phase commit, via a transactional outbox.

Production-ready by design

What decides success in operations.

Multi-AZ ReplicationmTLS & ACLsDead-Letter-QueuesBackpressure-HandlingConsumer-Lag-AlertingExactly-once SinksGDPR-compliant retentionTiered Storage
Our streaming services

From architecture to operations

architecture

Architecture & strategy

Event modelling, topic and partition design, delivery guarantees and the right platform, on-prem, cloud or hybrid. Technology-open, no vendor lock-in.

build

Implementation

Producers, consumers, Kafka Connect and stream processing with Kafka Streams & ksqlDB.

cloud_sync

Migration & integration

Connect existing systems to Kafka, CDC from SAP, Oracle and others, without a big-bang switch.

monitoring

Operations & monitoring

Managed operations with an SLA: monitoring, scaling, security and updates without downtime.

Streaming tech stack

We choose technology by requirement, not by vendor.

Apache KafkaConfluentApache PulsarApache FlinkKafka StreamsKafka ConnectksqlDBSchema RegistryDebezium CDCApache Avro / ProtobufAWS KinesisAzure Event Hubs

Related

UW
YOUR DIRECT CONTACT
Uwe Winnwa
Sales · cloud37 AG

30-min intro call — free.
No standard pitch. I listen, ask the right questions and give an honest assessment — even if we are not the right partner.