Alex SotisPortfolio

About Me

Alex Sotis

Background

I'm Alex Sotis — ten years a data practitioner, and a math nerd prior to that. Those years have covered internal business intelligence teams, external consulting firms, and product companies.

As a full stack data engineer, I work across the whole path the data takes: the infrastructure it runs on, the pipelines that bring it in, the models that shape it, and the semantic layer analysts and AI agents query at the end.

Currently a Staff Data Engineer at Fanatics, where I built out our medallion architecture, lead teams of data engineers and analytics engineers, and work alongside platform, AI, and backend engineers to keep the data stack coherent rather than stitched together.

Approach

With a strong understanding of business, I approach technology in economic terms: what a dataset costs to produce, and what the business actually does differently because it exists. Much of the job is translation, standing between the people who know the business and the people who know the systems so neither side has to guess what the other means. I consider myself a thought partner to stakeholders, and not merely someone who receives requirements.

On ingestion I build for failure. Pipelines should be idempotent and self-healing, so a bad night recovers on its own instead of waking someone up. I would rather spend the time on a framework than on another one-off DAG, so that adding a source becomes a config change a teammate can review in a minute rather than a script only its author understands.

The same goes for what comes after. I version control the data, not just the code that produces it, and keep automated tests in CI so bad code never reaches production and bad upstream data gets caught before a stakeholder finds it in a dashboard.

For modeling, columnar warehouses like Snowflake and BigQuery make the medallion architecture worth its extra layers. I start by writing the business process out in plain English and pulling the nouns out of it; those become the entity models, with one big table built on top. Analysts and AI agents get a single source of truth they can query quickly without writing joins.

The AI work comes out of the same thinking. Getting a new engineer productive on day one without tribal knowledge turns out to be the same problem as platforming an AI: written context and documentation, reusable scripts that make manual work fast and deterministic, and standard patterns written down where anyone can find them. Done well, the onboarding doc and the agent's context are the same artifact.

Skills

SnowflakeBigQuerydbtSQLPythonClaude CodeAI AgentsLookerTableauGitCIAirflowFiveTranGCPAWSTerraformMachine LearningBQML