Lead Data Quality Engineer
Posted Updated
Description
Responsibilities:
- Quality systems architecture: design and build the verification, sampling, and scoring pipelines that run continuously over billions of rows across every market.
- Frameworks: the team builds on the harnesses, abstractions, and SKILL. md specs that make new quality checks fast to write, cheap to run, and hard to get wrong.
- Harness engineering: evaluate harnesses for LLM-based validation and extraction: labelled eval sets, precision/recall tracking, judge calibration, prompt versioning, and drift detection on vendor model updates.
- Release gates: pre- and post-production gates that block bad data before it reaches customers, with failure analysis and triage tooling to match.
- Incident remediation: root-cause quality incidents at scale, ship the fix, and turn the failure into a permanent automated check.
- Team leadership: lead 3-5 applied AI engineers: technical direction, code review, pairing, and growing them into engineers who own outcomes end to end.
Requirements:
- 7-10 years building production data systems in business-critical environments: you've shipped systems that ran unattended, at scale, and stayed up.
- Worked with billions of rows; you know what breaks at that scale and how to design quality checks that don't.
- Built data quality systems and frameworks that were not used; built them: validation engines, anomaly detection, scoring, sampling strategies, and reconciliation against ground truth.
- Harness engineering experience: you've designed and evaluated or tested harnesses that other engineers depend on, and you can show the repo.
- Led engineers: You've directed a small team technically, reviewed their code, and stayed hands-on while doing it.
- Strong Python and advanced SQL, production-grade, performance-aware, and comfortable with concurrency and large-scale transformations.
- You operate LLMs as production systems, evaluate sets, version prompts, log traces, set cost ceilings, and debug judges on precision/recall.
- Shipped real work with agentic IDEs like Claude Code, Cursor, or equivalent. Not "tried it"; built and shipped with it as your default mode.
- Sharp judgment on rules vs. LLMs' deterministic checks where structure allows, LLMs where semantic judgment is needed, and you can defend the call.
Highly Valued:
- B2B data: firmographics, people data, entity resolution, registry matching across markets.
- Cloud data platforms: Snowflake, Databricks, Redshift, and AWS for pipeline deployment.
- Airflow (or equivalent) orchestration at production scale.
- Vector databases, embeddings, or retrieval patterns for matching and deduplication.
- Startup or scaleup experience where you defined the standard rather than inheriting it.
Skills
Sign in to read and join the discussion.