Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Data is not just the new oil; it’s the new language of decision-making. But before machine learning, embeddings, or even NoSQL, data had to be structured, described, and designed. Understanding how data models evolved, from Edgar Codd, Peter Chen and his ER diagrams, to neural vector representations is critical for any modern data practitioner. This post traces that journey.
1970 – Edgarr Codd
Edgarr Codd was working as a mathematical programmer for IBM when he wrote the paper “A Relational Model of Data for Large Shared Data Banks”. Thia paper, with its unassuming title, paved the way for a multi-billion pound industry. The concepts and terms used are still as relevant today as it was over half a century ago.
In the modern world of cloud computing, LLMs, agents and tooling like dbt, we often forget the decade old underpinnings that got us to where we are today. Concepts like logical modelling underpin this work and the evolutions that then occured.
I always ask the question “…who was Edgar Codd?” when hiring for data modellers. And a suprising number of people don’t know. Ask who “invented” the IPhone and you’d probably get an answer each time (the blog post on logical models above delves into why this is likely the case).
The paper Codd wrote introduces us to the concept of normal forms and the fact data can be represented as a relational model, rather than a hierarchy of trees. He expanded on the concepts of logical and physical models therein.
A new language was introduced that talked to these models in the form of relational calculus. Another pivotal idea was defining data using relations (tables), tuples (rows), and attributes (columns), with mathematically defined properties. This abstraction was new; data elements were no longer bound to storage paths or formats, but to positions in a relation.
We were introduced to the concept of keys to uniquely idetify rows and this was all underpinned by the concept of a large, sharable space of data used by multiple cuncurrent users.
