Make the platform self-serve: a registry of every dataset with its owner, schema, freshness and access; provisioning and permissions as code.
Run it in production: everything as Terraform and Helm with metrics, traces and alerts; be on call; handle incidents end to end, write the RCA and fix the root cause.
Grow the platform: new products keep arriving. You will help decide what the platform should do next and build it, whether that is a new source or sink, or a new way for teams to use the data.
WHAT WE'RE LOOKING FOR
2–5 years of experience building, shipping and running large-scale, data-intensive systems in production, ideally at a startup where you owned things and were on call for them.
Deep expertise in at least one of: an OLAP database (ClickHouse preferred; Druid, BigQuery or Redshift also fine) or Kafka / event streaming. You understand how a columnar store writes, merges and scans data, or how a partitioned log gives ordering and delivery guarantees.
Experience with large-scale messaging systems like Kafka or RabbitMQ: partitions, consumer groups, delivery guarantees, backpressure.
Good at Go, or good at Python.
You know the usual data-system problems: schema evolution, idempotency, backpressure, partitioning, retention and TTL, CDC.
You have built or run at least one of: an event pipeline, a transformation layer (dbt, Spark or SQL on Airflow), a provisioning or control-plane service, or a multi-tenant data service with access control.
Comfortable with AWS, Kubernetes and Terraform in production.
You test before shipping, debug with metrics and traces, write RCAs, and document what you build.
Passion for data infrastructure and for enabling others by making their data easier to access.
NICE TO HAVE
ClickHouse in production: MergeTree, materialized views, projections, quotas, settings profiles, row policies, ClickHouse Cloud.
Kafka in depth: MSK, Kafka Connect, schema registry (Glue or Confluent), Avro.
dbt and Airflow at scale.
Audit or security data systems: append-only logs, integrity checks, tenant isolation.
OpenTelemetry, LLM observability (Langfuse or similar), eval pipelines.
Built an internal platform that other teams adopted.