Services

Streaming & event sourcing.

Real-time streaming and event-sourced systems with Kafka, Redis and Postgres — an immutable log of events, projected into views you can trust. NZ-based.

Some systems can't wait for the nightly batch. When the truth is the sequence of events — orders, sensor readings, positions, ledger entries — we build streaming pipelines and event-sourced services that capture each fact as it happens, keep an immutable log of what occurred, and project that log into the read models your apps and dashboards actually query.

Where we help

Real-time ingestion and processing with Kafka and Redis streams
Event sourcing with an append-only log as the system of record
CQRS read models and projections rebuilt from the event history
Stream processing — windowing, joins, aggregations and enrichment
Exactly-once and idempotent consumers that tolerate replay
Backpressure, partitioning and ordering guarantees under load
Outbox patterns to publish events reliably from transactional writes
Live dashboards and alerting fed straight off the stream

Example builds

Live tracking platform. We built VesselTrack, which turns a continuous feed of positions into a live, queryable picture — the canonical case for treating a stream of events as the source of truth.

High-frequency price feed. Our FuelClock work taught us the shape of fast-moving, ever-changing data and how to keep it legible to people and systems downstream in real time.

Real-time inventory. In StockTrack every movement is an event; the current stock figure is just a projection of that history — replayable, auditable, and never quietly out of sync.

How we engage

We model the domain as events first, then derive state from them — which gives you a complete audit trail and the freedom to build new read models from the same history later. Kafka or Redis carries the stream, Postgres holds the projections, and consumers are idempotent so a replay is safe rather than scary. Single source, plural truths: one immutable log, many projections. See our services for how we engage.

Common questions

Do we really need event sourcing, or just a queue?: Often a queue and good projections are enough. Full event sourcing earns its keep when audit, time-travel or rebuilding new views from history genuinely matter. We are honest about where the simpler option is the right one.
How do you handle ordering and exactly-once delivery?: Partitioning to preserve order where it matters, idempotent consumers, and outbox or transactional patterns so events publish reliably with the database write. Effectively exactly-once processing without the folklore.
Can streaming sit alongside our batch pipelines?: Yes, and it usually should. Streaming handles the real-time view; batch and ELT handle heavy historical analysis. We connect both to the same warehouse so the numbers agree.

Data architecture ETL & ELT pipelines Business intelligence & BI

Considering this for a project?

Start a conversation

Where we help

Real-time ingestion and processing with Kafka and Redis streams
Event sourcing with an append-only log as the system of record
CQRS read models and projections rebuilt from the event history
Stream processing — windowing, joins, aggregations and enrichment
Exactly-once and idempotent consumers that tolerate replay
Backpressure, partitioning and ordering guarantees under load
Outbox patterns to publish events reliably from transactional writes
Live dashboards and alerting fed straight off the stream

Example builds

High-frequency price feed. Our FuelClock work taught us the shape of fast-moving, ever-changing data and how to keep it legible to people and systems downstream in real time.

Real-time inventory. In StockTrack every movement is an event; the current stock figure is just a projection of that history — replayable, auditable, and never quietly out of sync.

How we engage

Common questions

Do we really need event sourcing, or just a queue?

Often a queue and good projections are enough. Full event sourcing earns its keep when audit, time-travel or rebuilding new views from history genuinely matter. We are honest about where the simpler option is the right one.

How do you handle ordering and exactly-once delivery?

Partitioning to preserve order where it matters, idempotent consumers, and outbox or transactional patterns so events publish reliably with the database write. Effectively exactly-once processing without the folklore.

Can streaming sit alongside our batch pipelines?

Yes, and it usually should. Streaming handles the real-time view; batch and ELT handle heavy historical analysis. We connect both to the same warehouse so the numbers agree.