Why GRAM?

We have been in AI/ML for over 20 years, and when the world began looking at network graphs with Large Language Models (LLMs), we started envisioning something like an ORM (Object-Role Model) graph being ideal, because it captures precise physical structure of your database and deep semantic grammar for every table and data point.

A better semantic layer was definitely needed, because how can you trust an LLM to look at your database schema and simply guess what query to write? What about missing foreign keys? Cardinality? Uniqueness? Weird column names? Role-modeling dimensions?? For analytics, there seemed to be a lot of landmines for an LLM to navigate.

We tried all the big vendors' semantic layers and they lacked two key things a good analytical data model should have: simplicity above all, plus deep links to operational semantics, i.e. a conceptual view of how the data behaves and flows in the context of the business. How difficult could this be?

We needed a portable, standard-issue semantic layer.

Once you recognize that you cannot just hand your LLM an ERD and expect accurate resuts, then it is clear that better strategies are needed. Telling it to guess is not an option. So we went back to the basics: predicates, facts, relations, tuples, and other weird words that you don't typically think about much in the world of data because we expect our tools to just handle it.

And in doing so, we decoupled the semantic layer from the vendors. We put your grammar in your hands.

We bring to you Graph-Rule Analytical Mapping (GRAM). It is a new abstraction layer that brings database structure and business grammar together in a single semantic contract, and tools to bring it all to life in minutes. The AI doesn't infer your schema, it reads it.