Begin AI Documentation

Looking for some information on our terminology? Expand the headings here to get more information on our terminology.

Data Schema

A Data Schema is a JSON file used to define key data points within your software’s codebase. It contains the parameters used to generate instructions.

The data types are expressed as either:

text

date

category

boolean

number

image

video

Here’s an example of a schema based on a video and its title:


{
	"user":{}
	"product":{}
	"content":{
		"title":{
			"type":"text"
		}
		"video":{
			"type":"video"
		}
}
	"user_user_interactions":{}
	"user_content_interactions":{}
	"user_product_interactions":{}
}

Project

A Project is the combination of Begin’s machine learning algorithms and an uploaded data schema. Each project is tailored to a specific objective, enabling you to detect fake profiles while recommending a new local bakeshop to your users.

Instruction

An Instruction is how Begin’s machine learning systems communicate with devices for data processing. Instructions are generated using a selected schema once a project is created.

The SDK will apply these instructions to the end device, creating a profile of interactions and preferences (aka Signature) which are then used to learn and improve the end user's experience within your app.

💡

FOR EXAMPLE: A movie streaming app uploads a schema outlining its services through genre, sub-genre, cast, and theme. The base instructions in the app would generate a signature that outlines what kind of movies they like. That signature can then be used to recommend more movies.

As signatures develop, Begin’s learning algorithms will update instructions over time to further improve the user’s experience (in the above example “find a movie the end-user will watch” could turn to “find a movie the end-user will watch from start to finish.”)

Signature

Signatures (otherwise known as “embeddings”) are a profile, expressed as a mathematical hash meaningful only to the algorithm.

The signature is generated and referenced by the central server with every other signature in the app. Each individual signature is then improved and sent back to the device where the experience is tailored to the user.

💡

FOR EXAMPLE:

A movie streaming recommendation algorithm processes a user’s interactions. The user:

Viewed more movies under the “Science Fiction” genre

Viewed more movies with the tag “Strong Female Lead”

Reacted positively to movies tagged with “complex themes”

Reacted negatively to “gory”

This is referenced by the server, which references this with every other signature, and finds that users with a similar signature enjoyed Arrival. The signature is modified and the instructions are improved to recommend these movies. The process is repeated after the next round of interactions to further understand the user’s taste.

Key Terms