Skip to content

DataNotesData engineering

Data Architect Roadmap

Baraa Khatib Salkini8 min read

Subscribe to DataNotes

Join 213,000+ curious readers.

Subscribe takes you to Substack to confirm, and you can unsubscribe anytime (privacy policy).

Subscribe to DataNotes

Subscribe takes you to Substack to confirm, and you can unsubscribe anytime (privacy policy).

In this issue8 min left

Architecture isn’t about perfection. It’s about design, clarity, and intention.Over the years, I’ve learned that good data architecture has less to do with technology and more to do with thinking.Here’s a breakdown of the core skills every data architect must master :

1. Understand the Core Data Architectural Designs

There isn’t one single “data architecture.” Different designs solve different types of problems.

Key patterns to designs that you to learn:

  • OLTP vs OLAP
  • Data Warehouse → structured, clean data for analytics and BI.
  • Data Mart → smaller warehouse focused on one department or topic.
  • Data Lake → stores all data types (raw, semi-structured, structured).
  • Data Lakehouse → combines lake flexibility with warehouse reliability.
  • Lambda Architecture → combines batch and real-time processing.
  • Kappa Architecture → simplified real-time (stream-only) design.
  • Data Mesh → decentralizes data ownership by business domain.
  • Data Fabric → unifies data access across clouds and systems.
  • Data Product Architecture → treats each dataset as a reusable product.
  • Streaming Architecture → handles continuous, real-time data (Kafka, Flink).
  • API-Driven Architecture → exposes data through APIs for other systems.
  • Serverless Architecture → runs scalable data workloads on demand.
  • Data Virtualization → queries data across sources without moving it.

You do not need to build all of them.You only need to understand their purpose and trade-offs.

2. You “Design” Architectures

Architects draw. A lot. That’s how ideas turn into systems.Your job is to listen, ask questions, and sketch how data should flow.

You start from the user’s need, trace it back to the data source, and build the bridge in between.

What every architecture should include

  • Scalable and ready for future growth
  • Supports multiple use cases such as analytics, AI, and machine learning
  • Cost efficient and maintainable
  • Meets real business needs

One of the most common tools used by data architects today is Draw.io.It’s free, simple, and powerful enough for professional architecture design.

Study open reference architectures from Databricks ,Snowflake, or Azure Synapse.See how they structure ingestion, storage, and presentation layers.

3. You Can’t Architect What You’ve Never Built

If you want to design real systems, you first need to understand the world of data engineering.You need to know what it feels like to write SQL that breaks, to debug Python scripts, to manage GitHub conflicts, and to wait for a PySpark job that runs forever.

You cannot skip that stage.You can study the theory, but without building real systems, your designs will miss reality.You will end up creating architectures that look great on slides but fail in production.

That does not mean you will keep developing once you become an architect.And it does not mean you need to master every possible technology.But you should have built at least one or two complete data systems … a warehouse, a data lake..etc.

When you reach the architect level, your work shifts from building to designing.You will still create small proofs of concept, but most of your time will focus on shaping the overall picture and guiding others.

Essential tools: SQL, Python, GitHub, PySpark, dbt, KafkaCore concepts: ETL and ELT, Streaming, Batch, Data Pipelines

Spend a few years building. No book or course replaces that experience.

4. Development of PoCs

A big part of that work is developing Proofs of Concept, or PoCs.They are small, focused experiments that answer one question: Does this idea really work for us?

The pace of change is crazy. Every week new tools appear, and you are expected to know which ones fit your company’s data strategy.You move from “I think this might work” to “I know this works, and here’s the proof.”

How to stay sharp:

  • Keep reading release notes and official blogs from major platforms like Databricks, Snowflake, and Microsoft Fabric.
  • Follow new trends but stay critical. Learn to spot what is real progress and what is hype.
  • Track which tools are maturing and which ones are fading. Knowing when a technology is dying is as valuable as knowing what is new.

Never trust theory alone. Build small experiments until proof replaces opinion.

5. Data Modeling (Conceptual, Logical, Physical)

One of the most common things you will do as a data architect is model data.You will often be asked how to build an integrated data model that connects everything across the enterprise.

Data modeling may sound technical, but it is one of the most important communication tools you have.It turns complexity into clarity… the foundation every modern data architecture depends on.

To do that, you need to understand the main types of data models and how each one fits into real systems.

3 Levels of Data Modeling

  1. Conceptual Model: high-level business view of entities and relationships (no technical details).
  2. Logical Model: defines data structure with attributes, keys, and relationships — independent of any database.
  3. Physical Model: implements the logical model in an actual database (tables, columns, indexes).

Core techniques

  • Denormalization/Normalization → intentionally adding redundancy to speed up analytical queries.
  • 3NF (Third Normal Form) → design to eliminate redundancy in OLTP systems.
  • Dimensional Modeling → organizes data into facts (events) and dimensions (context) for analytics.
  • Slowly Changing Dimensions (SCD) → technique for tracking historical changes in dimension data.
  • Star Schema → one central fact table linked to multiple dimension tables.
  • Snowflake Schema → normalized version of star schema (dimensions split into sub-tables).
  • Data Vault → modeling method for large, scalable warehouses (hubs, links, satellites).

A good data model is not about diagrams. It is about making data simple, clear, and consistent for everyone.

6. Understand Platforms and Cloud Ecosystems

Every platform has its own philosophy and strengths.You don’t just need to know how to use them .. you need to know when to choose them.

Why it matters

Architecture isn’t only technical. It’s also strategic.Choosing the wrong platform can lock your company into costly limits.

How to learn it

Compare real projects across Databricks, Snowflake, Azure, AWS, and GCP.Study how each handles compute, storage, cost, and security.

Quick snapshot

  • Databricks: open, flexible, AI-ready
  • Snowflake: simple, fast, secure
  • Azure: enterprise-integrated
  • AWS: modular, powerful, but complex

A good data model is not about diagrams. It is about making data simple, clear, and consistent for everyone.

7. Create Standards and Governance

One of your main jobs as a data architect is to make sure everyone in the company speaks the same data language.That is what standards and governance are for.

If every team defines data differently, confusion spreads fast.Finance, sales, and operations all report different numbers, and no one knows which one to trust.Good governance fixes that by creating one shared version of truth.

As the architect, you design how this works.You define naming rules, ownership, and metadata so data stays consistent and easy to find.Your goal is to make sure every system connects smoothly and everyone understands what the data means.

Popular tools

Microsoft Purview, Collibra, and Alation are widely used in large companies.Smaller or cloud-native teams often use Atlan, OpenMetadata, or DataHub.For data access and quality, tools like Immuta, BigID, and Great Expectations help keep things secure and reliable.

Without standards and governance, data loses meaning, trust breaks, and every team ends up building its own version of truth.

8. Data Product Thinking and Data Mesh

A modern data architect designs data as products, not just pipelines.Each dataset has a clear purpose, an owner, and defined inputs and outputs.

This mindset helps large organizations grow without chaos.Every domain manages its own data but follows shared rules so everything stays consistent.That is the idea behind the Data Mesh.

How to learn it

Read Data Mesh by Zhamak Dehghani.Redesign one of your pipelines as a small data product with clear ownership and value.

Tools to know

Databricks Unity Catalog, Snowflake Marketplace, OpenMetadata, DataHub.

Treat data like a product, not a project. Products are owned, maintained, and improved over time.

9. Designing for AI and GenAI Workloads

Modern architectures must be ready for AI.Your job is to design systems that deliver clean, traceable, and reliable data for machine learning and generative AI.

AI fails when data quality is weak or lineage is missing.Good architecture keeps everything reproducible, from raw data to model training.

How to learn it

Start with the basics of MLOps and understand how data flows through ML pipelines.Then learn how metadata and lineage support repeatable results.

Tools to know

Databricks MLflow, Feature Store, Azure AI Studio, Vertex AI, Snowflake Cortex.

Leadership and Communication

And here is the most important skill.Architecture is not only about data. It is about people.As an architect, your main job is to bring clarity and help teams see how everything fits together and why it matters.

You will spend much of your time explaining, teaching, and guiding.Strong architects make complexity simple. They connect people as much as they connect systems.

Where this skill shows up

  • Leading project kickoffs and setting a shared vision
  • Reviewing architecture designs and guiding trade-offs
  • Aligning data, analytics, and ML teams on one strategy
  • Translating business goals into clear technical plans
  • Recommending tools and platforms that fit the data roadmap
  • Defining data standards, governance, and ownership
  • Solving performance and scalability challenges
  • Mentoring engineers and reviewing designs or code
  • Presenting architecture plans to leadership and stakeholders
  • Explaining complex data concepts in simple, human language

Final Thoughts

Becoming a great data architect is not about knowing every tool or buzzword.It is about understanding how systems connect, how people use them, and how decisions shape data for years to come.

  • You start with the fundamentals, the core architectural designs that define how data moves.
  • Then you learn to draw, to design systems that scale and make sense to others.You build before you architect, because real understanding comes from experience.
  • You test ideas through small proofs of concept, always learning what works and what does not.
  • You master data modeling, because it turns business complexity into something everyone can understand.
  • You explore cloud platforms, governance, and new approaches like data mesh so your designs stay modern and relevant.
  • You learn how to design for AI, because that is where most architectures are heading.
  • And finally, you learn to communicate, because architecture only works when people understand it.

The best architects are not the ones who draw the most complex diagrams.They are the ones who bring clarity, build trust, and make others better.

Thank you for being part of the Data With Baraa community. ❤️Your support makes it possible to create stuff like this!

Baraa