⚡ Netflix Ended Data Chaos with Unified Domain Models

knowledge graphs, automated schema generation, and domain-driven design

TL;DR

On June 12, 2025, Netflix unveiled UDA (Unified Data Architecture) – a foundational system that models business entities like ‘movie’ or ‘actor’ once and projects them consistently across APIs, databases, and pipelines. By combining knowledge graphs, automated schema generation, and domain-driven design, Netflix tackles the problem of fragmented data models that plague large-scale distributed systems.

---

Why does this matter?

For senior engineers managing data platforms, UDA solves a problem that emerges as organizations scale:

At Netflix, core concepts like ‘movie’ or ‘actor’ were redefined independently in each system, the GraphQL gateway powering internal tools, the media asset management platform storing content, the encoding pipeline transforming assets, and countless microservices. Each team modelled the same entities differently. The result? Inconsistent definitions, terminology drift, data quality issues, broken references, and brittle integrations that require heroic manual efforts to patch.

For any company with rapidly growing microservices and domain teams, these challenges are inevitable. UDA is Netflix’s approach to future-proofing data interoperability at scale.

---

What is UDA?

Netflix describes UDA as a “foundation for connected data”. Conceptually, it is a knowledge graph–driven system where:

1. Domain models are defined once – formal conceptualizations of business entities like movies, actors, or games.

2. These models are projected automatically into concrete schemas – GraphQL, Avro, SQL, Java – preserving their semantics everywhere they are used.

3. Mappings connect models to real data containers – like GraphQL resolvers, Data Mesh sources, or Iceberg tables, enabling systems to find and use data consistently.

4. Data moves faithfully between systems, with meaning preserved, without brittle manual ETL transformations.

5. Everything is organized as a knowledge graph, enabling search, introspection, and programmatic graph traversal to uncover relationships across systems.

---

Upper: The Metamodel Engine

At the heart of UDA lies Upper, Netflix’s metamodeling framework. It defines the language for domain modeling:

Upper itself is a bootstrapping metamodel – it models itself, defines what a domain model is, and validates its own definitions. This self-referencing design allows Netflix to generate:

• Type-safe Java APIs for domain models.

• GraphQL schemas federated into their Enterprise Gateway.

• Avro schemas for Data Mesh ingestion.

Because all models are conservative extensions of Upper, new domains integrate seamlessly into Netflix’s runtime and developer tooling.

---

How UDA Changes Data Engineering at Netflix