Our tech
The empirical layer is our name for where an organization keeps what it has learned from its data: how values are distributed, which columns move together, what drives an outcome, where the data has problems. Each finding is kept with the data it came from, so people and AI can open it, check it and build on it.
Analysts, dashboards, AI assistants and agents.
Distributions, relationships, effects and fitted models, each kept with the data behind it.
Warehouses, databases and files.
Metrics and business rules, like how revenue is defined.
With one, that knowledge lasts. When an analyst or an AI works something out, the finding is kept with its data, so the next question starts from there.
Your semantic layer holds what people decided, like how revenue is defined. The empirical layer builds on it, adding what the data actually shows on top of your data and your definitions. Semantic layers have an open standard, Apache Ossie ↗, and Elbi reads and writes it.
In SleuthBench, our study of how AI models answer questions about data, every leading model we tested answered what-drives-the-outcome questions more accurately when it had an empirical layer to work from. Read the paper ↗
For your organization: Summand builds one across your data, and it grows with every question your team asks.
For your own AI agent: Elbi, our open-source project, builds one on your machine. Its name says so: Elbi is short for Empirical Layer by Intelligible.