The data-engineering roadmap

From zero to data engineer — the whole map.

A free, honest reference for anyone building a data career: what the roles really are, a phase-by-phase roadmap, the 2026 modern data stack mapped out, the concepts and tools to master in each of nine tracks, and a curated reading list. Newbie or guru — start where you are.

Start here

However you got here, there's a path.

Completely new? Don't buy a course yet. Start at Foundations — three months of SQL and Python is worth more than any tool. Then walk the roadmap phase by phase and build the milestone project at the end of each.

Already working with data? Skim the stack map to see what's changed in 2026 (Iceberg, Dagster, dlt, DuckDB), then use the nine tracks as a checklist to find and close your gaps.

One rule holds for everyone: learn by building. Every track here maps to hands-on exercises on EgyBI — real, cited Egyptian data instead of toy tables. The books are pointers; the reps are the point.

The roles, decoded

Data engineer, analytics engineer, analyst — who owns what.

Three roles sit along one assembly line: raw data flows from the engineer, through the analytics engineer's models, to the analyst's answers. Know which one you're aiming for — the roadmap below serves all three, and they share the same foundations.

Upstream · builds the plumbing

Data Engineer

Moves and shapes data at scale: ingestion pipelines, storage, warehouses/lakehouses, streaming, and the reliability that keeps it all running in production.

Owns: pipelines, infrastructure, performance, uptime.
Core tools: SQL, Python, Spark, Airflow/Dagster, Kafka, a cloud (AWS/GCP/Azure), Docker.

~US$130k median · varies by market
Midstream · builds the models

Analytics Engineer

The bridge between engineering and analysis: turns raw warehouse tables into clean, tested, documented models the whole business can trust.

Owns: the transformation layer, metric definitions, data tests, docs.
Core tools: SQL, dbt, a cloud warehouse, Git, a semantic layer.

~US$150k+ median · fast-growing role
Downstream · answers the questions

Data / BI Analyst

Uses the models to answer business questions — dashboards, reports, and the story behind the numbers that drives decisions.

Owns: analysis, dashboards, KPIs, stakeholder insight.
Core tools: SQL, Excel, Power BI / Tableau / Looker, storytelling.

~US$105k median · the classic entry point

Adjacent roles share the same base: Data Scientist (stats + ML on top of clean data), ML / AI Engineer (models in production), and Platform / DataOps Engineer (the infra the data team runs on). Pay figures are US medians (Research.com / industry surveys, 2026) — adjust for your country; MENA and remote ranges differ.

The roadmap

Four phases from zero to job-ready.

A realistic order to learn things in — not a list of everything at once. Most career-changers reach job-ready in 6–9 months part-time (10–15 hrs/week), or 3–4 months full-time. Finish each phase with its project before moving on.

Phase 1

Foundations

The bedrock every data role shares. Don't rush it — depth here pays off forever.

Learn: SQL (joins → window functions), Python (data structures, files, APIs), the command line & Linux basics, Git, how data flows source → warehouse → dashboard.
Milestone: pull a real dataset, load it into a database, and answer 10 questions in SQL. Write a Python script that reads an API and saves clean CSVs.
≈ 0–3 months
Phase 2

Build the stack

Assemble a real, modern pipeline end to end — the core of the job.

Learn: dimensional modeling (star schemas, facts & dimensions, SCDs), a cloud warehouse (Snowflake or BigQuery), dbt for transformation, an orchestrator (Airflow or Dagster), Docker basics.
Milestone: API → raw storage → warehouse → dbt models (tested) → a dashboard, all scheduled and orchestrated. This is a portfolio project.
≈ 3–6 months
Phase 3

Production & scale

Make it big, reliable, and trustworthy — what separates juniors from engineers.

Learn: Spark/distributed processing, streaming (Kafka/Flink), the lakehouse & table formats (Iceberg/Delta), data quality & observability, CI/CD, cost & performance tuning, some infra (Docker → Kubernetes, Terraform basics).
Milestone: a scalable, tested, monitored pipeline handling both batch and streaming, with alerting when data breaks.
≈ 6–12 months
Phase 4

Specialize & deepen

Ongoing. Pick a lane and grow toward senior / staff.

Choose a lane: streaming & real-time · analytics engineering · platform / DataOps · ML/AI data engineering · governance & data contracts. Add system design, architecture trade-offs, and mentoring.
Milestone: design a data platform on paper, defend the trade-offs, and ship one hard thing others depend on.
Ongoing
Foundations (Phase 1)

The five things to nail before any tool.

Tools change every year; these don't. Get comfortable here and every framework afterwards is just syntax.

1 · SQL — deeply

The one language you'll use every day in every role. Go past SELECT: joins, GROUP BY/HAVING, subqueries, CTEs, window functions, and reading a query plan. Track 1 below is a full SQL path.

2 · Python

The glue of data engineering. Data structures, functions, files, error handling, and calling APIs. Then pandas/Polars for wrangling. You don't need to be a software architect — you need to be fluent.

3 · The command line & Linux

Pipelines run on servers, not laptops. Navigate a filesystem, pipe commands, ssh, edit files, read logs, manage environments. A weekend gets you dangerous; you'll use it forever.

4 · Git & version control

All modern data work is code — models, pipelines, configs — reviewed in pull requests. Learn clone / branch / commit / push / PR / merge. It's how teams ship without breaking each other.

5 · How data actually flows

The mental model: sources → ingestion → raw storage → transformation → warehouse/lakehouse → BI. Know why each layer exists before you learn its tools. The stack map below is exactly this.

Helpful, not blocking

Basic data structures & algorithms, a little networking (HTTP, ports), and spreadsheet fluency. You can pick these up as you go — don't let them stall you starting.

The modern data stack, 2026

Seven layers, and the tools that own them.

Every data platform is some arrangement of these layers. You don't need every tool — you need to understand what each layer does and one solid tool per layer. Named tools are today's common picks, not the only ones.

1 · Ingestion & extraction

Get data out of sources (databases, APIs, files, events) and land it — batch or streaming (CDC).

Batch/EL: Fivetran, Airbyte, dlt · Streaming/CDC: Kafka, Flink, Debezium

2 · Storage & the lakehouse

Where raw and modeled data live. Object storage + open table formats now blur "lake" and "warehouse" into the lakehouse.

Object store: S3, GCS, ADLS · Table formats: Apache Iceberg (the 2026 default), Delta Lake, Hudi, DuckLake

3 · Compute & query engines

The engines that actually crunch the data — from a single laptop to a thousand-node cluster.

Local/embedded: DuckDB, Polars · Distributed: Spark, Trino/Presto

4 · Warehousing

Managed, columnar, massively-parallel query platforms — the analytics workhorse.

Warehouses: Snowflake, BigQuery, Databricks, Redshift

5 · Transformation

Turn raw tables into clean, tested, modeled datasets — as version-controlled code.

Transform: dbt (table stakes), SQLMesh · plus a semantic layer for metrics

6 · Orchestration

Schedule, sequence, retry and monitor the whole pipeline — the conductor.

Orchestrators: Airflow (the incumbent), Dagster (asset-centric), Prefect, Kestra

7 · Serving, BI & the last mile

Where data becomes decisions — dashboards, reports, and self-service analytics.

BI: Power BI, Tableau, Looker, Metabase, Superset

Across it all · quality & governance

Trust is a layer too: testing, observability, lineage, cataloging and access control run across everything above.

Quality: dbt tests, Great Expectations, Soda, data contracts · Catalog: DataHub, OpenMetadata, Unity Catalog

The nine tracks in detail

Nine skills, grouped into three specializations.

Each track lists what it is, why it matters, the concepts to master (your checklist), the tools, a book pointer, and the hands-on EgyBI exercises — all on real, cited Egyptian data. They map to the three roles from the top of the page, so you can follow one path or borrow across all three. SQL underpins every path — start there whatever your goal.

Specialization · Data engineering

Build and run the pipelines.

For the engineer who moves and processes data at scale — ingestion, orchestration, distributed compute, and the cloud platforms it all runs on. Maps to the Data Engineer role.

Intermediate → Advanced · Phase 2

3 · Orchestration (Airflow / Dagster)

Pipelines aren't scripts you run by hand — they're scheduled, retried, and monitored. Orchestration is the conductor.

Concepts to master
  • DAGs, tasks & operators
  • Scheduling & backfills
  • Idempotency & retries
  • Sensors & triggers
  • Passing data between tasks
  • Dynamic pipelines
  • Asset-centric (Dagster) vs task DAGs
  • SLAs & alerting

Tools: Apache Airflow, Dagster, Prefect, Kestra.

Read: Data Pipelines with Apache Airflow (Harenslak et al.).

Practice on EgyBI
  • Live Ingestion DAG — pull the Met Museum's real nightly Egyptian-art CSV, land it, trigger transforms.
  • Backfill — recompute dynasty-level aggregates after a schema change.
  • SLA Alerting — fire an alert if the ingestion DAG misses its window.
Advanced · Phase 3

5 · Spark / PySpark & Distributed Data

When data outgrows one machine, you go distributed. Spark is the batch engine behind most large-scale pipelines.

Concepts to master
  • RDD vs DataFrame vs SQL
  • Lazy evaluation, transforms vs actions
  • Partitioning & the shuffle
  • Data skew & how to fix it
  • Broadcast joins
  • Caching / persist
  • The Catalyst optimizer
  • Structured streaming & Parquet

Tools: Spark/PySpark, Databricks, Parquet, Polars (single-node).

Read: Learning Spark (Damji et al.) · Designing Data-Intensive Applications (Kleppmann).

Practice on EgyBI
  • Partition Skew — partition the ~30k-row Met dataset by era; detect and fix skew.
  • Broadcast Join — broadcast the tiny dim_dynasty against a large fact table.
  • Idempotent Ingestion — exactly-once load of the Deir el-Medina records.
Advanced · Phase 3

6 · Cloud Warehouses & Lakehouse

The managed platforms where modern analytics lives. Understanding how they store and bill data is what makes pipelines fast and cheap.

Concepts to master
  • Columnar storage & MPP
  • Storage/compute separation
  • Partitioning & clustering
  • Micro-partitions & pruning
  • Time travel
  • Semi-structured (VARIANT/JSON)
  • Cost & credit models
  • Iceberg/Delta table formats

Tools: Snowflake, BigQuery, Databricks; Apache Iceberg.

Read: Snowflake: The Definitive Guide (Avila).

Practice on EgyBI
  • Clustering Key — choose and justify a clustering key for the Met CSV.
  • Semi-structured JSON — query the Met API's nested artist/constituent fields.
  • Time Travel — recover from a bad UPDATE to a monument-attribution field.
Beginner → Advanced · Phase 1–3

9 · Python for Data Engineers

The language that ties the whole stack together — ingestion, glue, automation, and the pipelines the orchestrator runs.

Concepts to master
  • Data structures & comprehensions
  • pandas / Polars
  • requests & working with APIs
  • Error handling, retries, backoff
  • Logging & typing
  • venv / packaging
  • Testing with pytest
  • CSV / JSON / Parquet; dlt for EL

Tools: Python 3, pandas/Polars, requests, dlt, pytest, SQLAlchemy.

Read: Data Engineering with Python (Crickard) · Fundamentals of Data Engineering (Reis & Housley).

Practice on EgyBI
  • Resilient Ingestion — a retry/backoff script against the real Met Museum API.
  • Clean the Papyrus — clean a Wilbour-Papyrus-style ~2,800-row land/tax dataset.
  • End-to-End Pipeline — ingest → land → dbt transform → orchestrate → BI-ready mart.
Specialization · Analytics engineering

Turn raw data into trusted models.

For the builder in the middle — dimensional modeling, dbt transformations, and the tests and observability that make the numbers trustworthy. Maps to the Analytics Engineer role.

Intermediate → Advanced · Phase 2

2 · Dimensional Modeling & Warehousing

How you shape data so a business can actually use it. The design skill behind every fast, trustworthy warehouse.

Concepts to master
  • Star vs snowflake schemas
  • Facts & dimensions
  • Grain of a fact table
  • SCD types 0–6
  • Surrogate keys
  • Conformed dimensions
  • Kimball vs Inmon vs Data Vault
  • Wide tables / OBT trade-offs

Tools: any warehouse, dbt, an ERD tool.

Read: The Data Warehouse Toolkit (Kimball) · Building the Data Warehouse (Inmon) · Star Schema (Adamson).

Practice on EgyBI
  • Monument Star Schema — build fact_monument_construction against pharaoh, dynasty & location dimensions.
  • SCD Type 2 — model the Great Sphinx's disputed attribution as slowly-changing history.
  • Conformed Dimensions — share dim_dynasty across monument and reign facts.
Intermediate · Phase 2

4 · dbt & Analytics Engineering

The modern transformation layer — SQL, but version-controlled, tested, documented and modular. The heart of the analytics-engineer role.

Concepts to master
  • staging → intermediate → mart
  • Materializations (view/table/incremental)
  • Generic & singular tests
  • Sources & freshness
  • refs, the DAG & lineage
  • Macros & Jinja
  • Snapshots (SCD)
  • Docs, exposures & the semantic layer

Tools: dbt (Core/Cloud), a warehouse, Git.

Read: Analytics Engineering with SQL and dbt (Machado & Russa).

Practice on EgyBI
  • Model Layers — build staging → intermediate → mart on the real Met dataset.
  • Test Suite — pass not_null / unique / relationships tests.
  • Incremental Model — process only Met objects updated since the last run.
Intermediate → Advanced · Phase 3

8 · Data Quality, Testing & Observability

A pipeline that silently ships wrong numbers is worse than one that's down. Trust is engineered — this is how.

Concepts to master
  • Quality dimensions (accuracy, freshness…)
  • Tests & assertions
  • Data contracts
  • Freshness / volume / schema monitors
  • Distribution & anomaly detection
  • Lineage & impact analysis
  • Incident management
  • SLAs / SLOs for data

Tools: dbt tests, Great Expectations, Soda, Monte Carlo.

Read: Data Quality Fundamentals (Moses, Gavish & Vorwerck).

Practice on EgyBI
  • Overlap Validation — no two same-dynasty pharaohs illogically overlap (co-regency allowlist).
  • Freshness Check — alert if the Met's nightly CSV hasn't updated in 48h.
  • Lineage — when a disputed date shifts, trace which marts need reprocessing.
Specialization · BI development

Query the data, turn it into decisions.

For the analyst and BI developer — SQL is the foundation every path starts with, and here it is most of the job, paired with the visualization skills that turn a model into something a business acts on. Maps to the Data / BI Analyst and BI Developer roles.

Beginner → Advanced · Phase 1

1 · SQL — Fundamentals to Advanced

The single most-used skill in data. Every role queries; the good ones query well. Master this before anything else.

Concepts to master
  • All join types & anti-joins
  • GROUP BY, HAVING, aggregates
  • Subqueries & CTEs
  • Window functions
  • Indexes & the query plan
  • Normalization & keys
  • NULL logic & set operations
  • Date/time & string handling

Tools: PostgreSQL, any warehouse SQL, SQLite (in-browser here).

Read: SQL Antipatterns (Karwin) · SQL Performance Explained (Winand).

Practice on EgyBI
  • Longest Reigns — rank pharaohs by reign length; a window-function variant partitions by dynasty.
  • The Family Line — self-join to reconstruct father→son succession chains.
  • Missing Monuments — anti-join UNESCO sites against monuments to find the gaps.
Beginner → Intermediate · any phase

7 · BI & Data Visualization

Data only matters when someone acts on it. BI turns models into dashboards, measures and stories a business runs on.

Concepts to master
  • Semantic models & measures (DAX)
  • Star schemas for BI
  • Choosing the right chart
  • Dashboards vs reports
  • Pre-attentive attributes
  • Row-level security
  • Self-service & governance
  • KPI & metric design

Tools: Power BI, Tableau, Looker, Metabase.

Read: Storytelling with Data (Knaflic).

Practice on EgyBI
  • Construction Gantt — a timeline of monument-construction eras across dynasties.
  • Flood Anomalies — visualize the Nile flood record, highlighting outlier years.
  • Fix the Chart — redesign a misleading dual-axis "monuments vs reign length" chart.
Core architecture concepts

The ideas behind the tools.

Tools come and go; these concepts are what interviews probe and production demands. Understand the why and any new tool is easy.

Batch vs streaming

Process data in scheduled chunks, or continuously as it arrives. Most teams run both — batch for history, streaming for now. Know the latency/cost trade-off.

ETL vs ELT

Transform before loading, or load raw and transform inside the warehouse. Cheap cloud compute made ELT the modern default — the L before the T is why dbt exists.

Warehouse · lake · lakehouse

Structured & fast, cheap & flexible, or both at once. Open table formats (Iceberg) gave the lake warehouse-grade reliability — the 2026 direction.

Idempotency

Running a pipeline twice must not double the data. Idempotent loads (upserts, partition overwrites) are what make retries and backfills safe.

Data contracts

A promised schema & meaning between a producer and its consumers, so an upstream change can't silently break downstream models. The 2026 answer to "who broke the dashboard?"

Partitioning & the shuffle

How data is split across files/nodes decides speed and cost. Moving it around (the shuffle) is expensive — designing to avoid it is half of performance tuning.

The reading list

The shelf worth building.

A short, respected library that covers the whole field. Read the first two cover to cover; use the rest as track references. (Titles & authors only — EgyBI's exercises are written from scratch, never copied.)

Fundamentals of Data Engineering — Reis & Housley. The best single starting map of the whole field and its lifecycle.
Designing Data-Intensive Applications — Kleppmann. The deep "why" behind storage, distribution and reliability. A career-long reference.
The Data Warehouse Toolkit — Kimball & Ross. The canon of dimensional modeling — facts, dimensions, star schemas.
SQL Antipatterns — Karwin. Learn great SQL by studying the traps to avoid.
SQL Performance Explained — Winand. How indexes and the query planner really work.
Analytics Engineering with SQL and dbt — Machado & Russa. The modern transformation-layer workflow.
Data Pipelines with Apache Airflow — Harenslak et al. Orchestration from first DAG to production patterns.
Learning Spark — Damji et al. Distributed processing with the DataFrame API.
Data Quality Fundamentals — Moses, Gavish & Vorwerck. Observability, testing and trust at scale.
Storytelling with Data — Knaflic. Make the last mile — charts people actually understand.

Free resources worth more than most courses: the official docs for dbt, Airflow/Dagster, Snowflake/BigQuery and Spark; open datasets (World Bank, the Met Museum's CC0 collection, and the cited data right here on EgyBI); and the data communities on Reddit, Discord and LinkedIn where practitioners debate this stuff daily.

Straight answers

Questions every beginner asks.

Do I need a computer-science degree?

No. A degree helps but isn't required — a strong portfolio of real pipelines and solid SQL/Python get interviews. Most hiring is skills-first now.

Do I need to know maths / statistics?

For data engineering, not much beyond comfort with numbers — that's more the data-scientist's world. You need logic, systems thinking and code. Stats matter if you later move toward data science.

Which should I learn first — Python or SQL?

SQL. It's the fastest to become useful with and every role uses it daily. Add Python right after; the two together are your foundation.

Cloud — which one, and do I need to pay?

Any one of AWS / GCP / Azure teaches the concepts; they transfer. Free tiers plus tools that run locally (DuckDB, Postgres in Docker) let you learn the whole stack for almost nothing.

How do I get experience with no job?

Build projects that look like real work: pull from an API, store raw files, transform with dbt, load to a warehouse, feed a dashboard — orchestrated and on a schedule. One end-to-end project beats ten tutorials. EgyBI's tracks are built to be exactly that.

Is AI going to replace data engineers?

It's changing the job, not deleting it — AI writes boilerplate, but someone still designs, tests, secures and runs the systems, and demand for clean, reliable data only grows as AI does. Learn to use the tools; own the judgment.

Stop reading. Start querying.

The roadmap only works if you build. Every track here has hands-on reps waiting — real, cited Egyptian data, auto-graded, in your browser.