Data Scientist / Lead Data Engineer

Senior Data Engineer / Data Platform Lead

The Role

Every number a customer argues about.

Architecture is still open and yours to decide and defend: warehouse shape, table format, orchestration, where the streaming boundary sits against change capture or a tighter batch cadence, and how datasets are tiered with real service levels behind them.

This is not an Analytics Engineer role. They decide what a term means; you decide whether the number is computed correctly, arrives on time, and can bear the inference drawn from it.

If a metric means two different things in two places, that is theirs. If it means the right thing but is stale or irreproducible, it is yours.

What You Would Actually Be Doing

Own the Event Pipeline

Own the event pipeline behind shipment milestones: late arrivals, corrections that restate a milestone recorded three days ago, and the gap between event time and ingest time.

Define how a restatement propagates to something a customer has already seen.

Build for Agent-Grade Serving

Make the serving path good enough for an agent to depend on — with a stated freshness target per dataset and a read path whose p99 you know.

Because when a run stalls waiting for a number, a workflow with money attached stalls too.

Make Data Quality Fail, Not Warn

Build data quality as failing checks, not dashboards:

• Null rates
• Referential integrity breaks
• Duplicate milestones
• Distribution shifts when a partner changes a field

A check that only warns is a check nobody reads.

Enforce Tenant Isolation

Enforce tenant isolation at the data layer and assert it through tests.

Contract rates are commercially sensitive between competitors who may share the same forwarders.

Own Expensive Backfills

Plan backfills that cost real money.

A reprocess across 30+ countries of historical data has a bill and a blast radius. You own both.

Apply Statistical Judgement

Bring statistical rigour to decisions.

Know what you can and cannot claim from observational data — and say so before the decision, not after it.

What We Look For

Must-Haves

Years are a floor, not the bar.

If you miss one line but are strong on the rest, apply and tell us which one.

• 5+ years building and operating production data platforms, including at least 2 years as the person accountable for architecture others depended on.
• Designed a warehouse from first principles — grain, dimensional models, slowly changing dimensions, additive vs. non-additive measures — and can describe a modelling trap you found in someone else's work and fixed.
• Deep SQL and Python: window functions, incremental and idempotent transformations, safe-to-rerun backfills, and pipelines that survive late-arriving data.
• Hands-on ownership of a production orchestrator — Airflow, Dagster, Prefect, Temporal, or equivalent — including retries, service levels, dependency modelling, and backfill strategy.
• A defensible position on table formats and streaming vs. change capture vs. tighter batch cadence, with the reasoning rather than the fashion.
• Genuine statistical rigour: can design an experiment, defend the power calculation, and pick the right test for skewed or correlated data.
• Has led engineers — hiring, reviews, mentoring — and carried a pager for datasets other teams depended on.

Also Good — None Required

• You have run a data on-call rotation and can describe what you changed to make it quieter.
• You have worked in a domain where being wrong about a time or quantity had a cost someone could name.

By Month Six

What Good Looks Like

We would rather tell you now what we would be measuring, so you can decide whether this is the job you want.

01 — Dataset Reliability

Every dataset a workflow depends on has a stated freshness target and a failing check behind it.

02 — Tenant Isolation

Tenant isolation in the data layer is asserted by a test that runs on every change.

03 — Team & On-Call Impact

You have hired or levelled up at least one engineer, and the data on-call is quieter than when you arrived.