Data Models: Harmonizing AI and Teams to Avoid Errors

⚡

Key Takeaways

1The importance of a common language is crucial for integrating AI into a team, as demonstrated by a startup in 2020.

2The data model, inspired by Google's 2017 article, transforms the way AI processes information by using attention.

3To avoid confusion, it is essential to clearly define the entities and their relationships in a system, as in a food delivery application.

💡Why it matters — A well-structured data model ensures that the AI and the team share a common understanding, avoiding costly mistakes.

A Necessity for Agreement

Before diving into the use of an artificial intelligence tool, it is crucial for your team to agree on the language to be used. In 2020, while I was working at a startup, our Chief Technology Officer realized that it was imperative to strengthen our technical skills to remain competitive. To achieve this, he hired a data scientist and encouraged us to take our work seriously. This process involved immersing ourselves in mathematics, using the pandas Python library, and conducting in-depth analyses of complex diagrams.

The Importance of the Data Model

A significant document I received at that time was titled "Attention Is All You Need," published in 2017 by a team of researchers at Google. This paper introduced the transformer architecture, which underpins every language model you have used. What struck me was not so much the mathematics but the concept itself. Prior to this, AI processed information sequentially, often losing track by the end of a sentence. The attention mechanism changed that, allowing the model to consider everything at once and learn what to focus on.

The Confusion of Words

If attention determines what is important, vectors and clusters are where that attention is actually directed. Instead of storing words as definitions, the model represents them as points in space. Words with related meanings end up close to each other. This means that if your team refers to the same thing by three different names, the model is not confused because it is dumb, but because you are.

The Impact of AI on Communication

When AI is involved, language gaps cease to be a coordination issue and become a product issue. The model does not ask for clarifications. It simply acts according to the definition it has inferred, confidently and at scale. To ensure that the model operates with the same definitions as your team, it is essential to create a data model.

Building a Data Model

Before building anything, it is crucial to name things, that is, the entities. For example, let’s imagine we are building a food delivery application. An entity is anything in your system that has a name, a definition, and attributes that describe it.

Here’s what we need:

User: the person ordering food, with a location, preferences, and order history.
Restaurant: the place preparing the food, with a menu, hours, and a location.
Menu Item: a specific dish or product offered by a restaurant, with a name, price, and a link to a restaurant.
Order: the transaction linking a user to a restaurant, with a status, a list of items, a total, and a timestamp.
Delivery Person: the individual delivering the order, with a location, availability, and an assigned order.
Delivery: the physical act of handing over the restaurant order to the user, with a route, status, and estimated time.
Payment: the financial record of the transaction, linked to an order and having a status.
Review: feedback left by the user after delivery, referring to the order, the restaurant, and sometimes the delivery person.

Grouping Entities

These entities do not all live in the same neighborhood. Let’s group them:

People and Identity: User, Delivery Person. These are the humans in the system, each with their own context, permissions, and goals.
The Offer: Restaurant, Menu Item. What is available, what can be ordered.
The Transaction: Order, Payment. The moment something happens, where money is exchanged.
The Experience: Delivery, Review. What the user actually feels.

Establishing Explicit Relationships

It is essential to clearly establish the relationships between these entities. For example:

What does it produce? A Restaurant produces Menu Items. An Order produces a Payment.
What does it reference? An Order references a User, a Restaurant, and one or more Menu Items.
What does it influence? User preferences influence which Restaurants appear first.

These three questions are your connectors. By using them, you can create a data model that helps align your team and your AI, ensuring that everyone works with the same understanding.