Learn what production data teams
actually do.
31 skills, 7 tracks, one clear path from “what’s a DAG” to “staff platform engineer.” Every skill ends with a deployable project.
How tracks build on each other.
Take tracks in any order — but most land a job fastest starting at Foundations → Batch → one specialty.
Your first three skills.
SQL Mastery for Data Engineers
Window functions, CTEs, execution plans — SQL you'd pass at a FAANG screen.
Python for Data Engineers
Production Python: typing, async, Pydantic, packaging — not a notebook tour.
dbt & Analytics Engineering
Incremental models, tests, macros, exposures — done the way prod requires.
The 32 skills production data teams actually use in 2026.
Most data-engineering roadmaps you’ll find online were written for the 2019 stack: a video on Spark, a video on Airflow, a video on Snowflake, done. The 2026 stack is materially different — and the curriculum below is built around what working production teams actually deploy this year, not what was fashionable five years ago.
The 7 tracks, and why the order matters
The 32 curricula are organized into 7 tracks: Foundations (SQL + Python + Data Modeling + Cloud), Data Systems (dbt + Spark + Airflow + Iceberg + Kafka + Flink), Quality + Governance (Observability + Contracts + Cost Optimization), Platform Engineering (CI/CD + IaC), Analytics (Metrics layer + Product Thinking), AI Systems (RAG + Vector DBs + LLM Pipeline + Agentic + Inference Serving + Feature Stores + MLOps + Dataset Engineering + Enterprise AI), and Leadership (Staff DE + System Design).
The recommended order is Foundations → Data Systems → either Analytics or AI Systems (depending on your target role) → Platform + Leadership. The reason: production AI systems sit on top of production data systems, and production data systems sit on top of SQL, Python, and dimensional modeling. Skipping foundations is the single most common reason engineers fail the senior-level system design round even when they can write Flink code.
Which curriculum should you start with?
If you can’t write a window function from memory, start with SQL Mastery (free). If you can write SQL but the words “Polars,” “Pydantic,” or “async I/O” are unfamiliar, start with Python for Data Engineers (free). If you have both but have never built a dbt project, start with dbt & Analytics Engineering. If you’re a working DE moving into AI, start with RAG → LLM Evaluation → Agentic Systems.
What “production-grade” means here
Every curriculum ends with a deployable artifact, not a quiz. The Spark curriculum ends with a Kubernetes-deployed PySpark job running with the Spark Operator + Prometheus metrics. The Airflow curriculum ends with a multi-DAG repo on the KubernetesExecutor with CI/CD and DAG-library versioning. The RAG curriculum ends with a hybrid retrieval system (BM25 + dense + RRF + cross-encoder reranker) benchmarked on recall@10. The shape is the same across all 32: read the ADR, deploy the system, break it, fix it, understand why the architectural decision went the way it did.
How the curriculum maps to interviews
Senior+ data-engineering interviews test three things: can you scope a problem before you draw boxes, can you defend a tradeoff on data you have, and can you talk about failure modes — not just the happy path. Every curriculum here is built around those behaviors. The system-design modules name the clarifying questions seniors ask. The cost-optimization modules teach the “reduce scan first, then add compute” ordering. The observability + governance modules teach what happens when a pipeline silently stops at 3am. That’s the gap between “I’ve done a Spark tutorial” and “I can defend a $300K Snowflake bill in front of finance.”
Below is the full catalog. Filter by track, tier, or hours to find your starting point.
Foundations
SQL Mastery for Data Engineers
Window functions, CTEs, execution plans — SQL you'd pass at a FAANG screen.
Python for Data Engineers
Production Python: typing, async, Pydantic, packaging — not a notebook tour.
Advanced Data Modeling
Dimensional, Data Vault, OBT — and when to reach for which.
Cloud Data Infra & FinOps
AWS/GCP for data teams — IAM, networking, cost, the parts that matter.
dbt & Analytics Engineering
Incremental models, tests, macros, exposures — done the way prod requires.
Batch pipelines
Apache Spark Deep Dive
Partitioning, shuffles, AQE, Iceberg — Spark for engineers who own the cluster.
Apache Airflow
Production orchestration: DAG patterns, idempotency, KubernetesExecutor.
Apache Iceberg & Lakehouse
Time travel, hidden partitioning, MERGE — the lakehouse format that won.
Warehouse Internals
MPP engines from the inside: query planning, distribution, spills.
Streaming
Real-Time Streaming Architecture
When to actually reach for streaming — and when you absolutely should not.
Kafka Streams
Topics, partitions, retention, consumer groups — Kafka with correct intuition.
Flink & Stream Processing
Stateful streaming, windows, watermarks, exactly-once with idempotent sinks.
Event Design & Contracts
Schemas that evolve safely. Protobuf + registry + CI compat tests.
Data quality
Data Observability & Quality
Contracts, GE, anomaly detection, lineage — and what's worth alerting on.
Governance & Data Contracts
Producer-side contracts, PII, access control, audit trails that hold up.
AI & vectors
RAG Learning Path
Hybrid retrieval, reranking, eval — production RAG, not a demo.
Vector Databases
pgvector, Qdrant, Lance — when to use each, and how indexing actually works.
LLM Data Pipelines
Batch enrichment with structured outputs, retries, cost budgets, at scale.
LLM Evaluation
Build a labeled eval set, run recall@k + LLM-judge, iterate without vibes.
Feature Stores for ML
Feast, point-in-time correctness, offline/online parity — no training/serving skew.
Agentic Workflows
Multi-step agent pipelines, tool use, trace eval, when to actually use them.
AI Inference & Serving
Online serving, model routing, autoscale, latency budgets in production.
Dataset Engineering
Curate, clean, version, and version-pin datasets that survive review.
Enterprise GenAI & Security
Multi-tenant LLM, prompt injection, secrets, audit — the hard parts.
MLOps for Data Engineers
CI/CD for models, monitoring, retraining, drift — the parts DEs actually own.
Platform
DataOps: CI/CD & IaC
Terraform, GitHub Actions, Argo, dbt CI — pipeline platform plumbing.
Cost Optimization for DEs
Attribute, alert, and reduce. The boring work that funds the fun work.
API & External Integration
Auth, retries, rate limits, idempotency — the unglamorous integration tier.
Staff Engineering
Product Thinking for DEs
Metrics, semantic layers, user-of-data thinking — bridge to the business.
System Design for DEs
Trade-off framework for the design round — and the architecture review.
Staff Engineer Playbook
Architecture, RFCs, calibration — the non-coding parts of staff DE work.
Not sure which skill to open first? We’ll pick.
Two minutes. Five questions. A sequenced learning plan and your free first lesson in your inbox.