Who & why
Turns data into production ML systems and reliable pipelines. Not "train a model in a notebook" (closer to Data Scientist) but building an engineering system: clean, timely data (data engineering) and models trained, deployed and monitored in prod (ML engineering / MLOps). Without one: pipelines break silently, models live "by hand", no drift monitoring, irreproducible training, unstable ML features.
A day in the life
Morning: check pipeline health & data quality, watch prod model metrics for drift. Day: build/fix an ETL transform, feature engineering, set up reproducible training & deploy, mind scale & cost. Evening: configure drift monitoring & quality alerts, write a model card, sync data architecture with the architect.
Key skills
Hard: Python, SQL, data processing (pandas/Spark), pipelines (Airflow/dbt), warehouses, feature store, ML basics, MLOps (data/model versioning, deploy, drift monitoring), containers/cloud. Soft: engineering discipline (reproducibility, data tests), data-quality focus, systems thinking, accuracy/cost/speed balance.
Artifacts
Data pipelines, model & quality metrics, feature store, drift monitoring, model card. Designs data architecture with the architect; done = DoD.
How AI / vibe-coding boosts the role
Pipeline generation; feature engineering; metric/drift triage; model card; cost optimization — with ready prompts.
Growth: Junior → Middle → Senior → Lead
Junior: transforms & simple pipelines. Middle: builds pipelines & model deploy, monitoring. Senior: designs the data/ML platform & MLOps. Lead/Head: data/ML infra & practices company-wide.
Common mistakes
"Works in a notebook" = done; target leakage; no drift monitoring; irreproducible training; garbage in.
What to learn
ETL/ELT, data modeling, feature store, ML metrics, MLOps (model CI/CD, versioning, drift monitoring), data leakage, scale/cost. Read: Designing Machine Learning Systems (Huyen); Designing Data-Intensive Applications.
Salary (RU)
Junior ~120–190k₽/mo, Middle ~190–330k, Senior ~330–550k+. High-paid niche; varies — check current data.
Laskoff agent mapping
No direct mapsTo; implementation by the engineer agent (data/ML), data architecture with the architect agent. Reproducibility & backup-before-irreversible map to the house rule + guard hook.
🤖 Persona prompt
You are an experienced Data/ML Engineer (data engineering + MLOps). Help me take data and models to prod as an engineering system, not a notebook. Make pipelines reproducible with data-quality tests (garbage in → garbage out). Compute features the same in training and prod; actively hunt target leakage. Version models and always set drift & quality monitoring — degradation must not pass silently. Mind compute scale and cost. On request give pipelines (dbt/Airflow), feature engineering, drift triage and model cards.