Insights from a decade of metadata-driven data access
Over the last ten years at Hasura, we’ve been focused on one mission: making enterprise data access fast, secure, and effortless.
We didn’t set out to build what the market now calls a “semantic layer.” Instead, our work on simplifying API development through a metadata-first approach naturally evolved into something that meets – and often exceeds – that definition.
The path to the unified semantic layer
From day one, our aim was to give teams a declarative, low-code way to create data access APIs. What data access could be provisioning based on a configuration?
That meant introducing a metadata blueprint that could captures:
- Business terms and logic as reusable, versioned building blocks
- Schemas, relationships, permissions, and constraints in a structured YAML format
- Auto-generated metadata through source introspection – databases, API, etc
- Version control and CI/CD alignment for enterprise workflows
The resulting artifact was a unified semantic supergraph – a single, authoritative definition of how data should be understood, secured, and accessed.
The distributed query engine behind it
The supergraph is run by our distributed query engine, which:
- Compiles incoming requests into the appropriate GraphQL, SQL, or API calls — applying filtering, access control, caching, and other rules at execution time
- Pushes queries down to the source systems and federates the results back
- Monitors and tracks usage, errors, and performance metrics at a granular level
Enterprises end up with a single, unified data access layer — straightforward to implement, simple to manage, and easy to monitor and evolve.
When AI entered the picture
By the time LLMs became viable for analytics, we already had a critical advantage:
A machine-readable, operationally enforced way to represent deep business context
The last step was teaching LLMs to use this artifact to reason about your business like a veteran analyst or engineer would.
That’s what led to PromptQL, which now delivers highly accurate AI-driven analytics and automations for some of the largest organizations in the world.
Read the ultimate guide to semantic layer for AI →
The core lesson
If you want AI-powered analytics to actually move the needle for the business, traditional semantic layer tech is not going to cut it. You need something that is:
- Quick to stand up with your existing definitions and sources
- Low-friction to keep current
- Close feedback loop so usage informs updates to improve accuracy
Anything less, and you’re left with little more than a static glossary — not a living, operational semantic layer.
Discover how you can drive reliable AI decision intelligence and automation with PromptQL → request demo
