Key concepts of Amazon Q Business

This section describes the key concepts and terms related to Amazon Q Business.

Retrieval Augmented Generation

Retrieval Augmented Generation (RAG) is a natural language processing (NLP) technique. Using RAG, generative artificial intelligence (generative AI) is conditioned on specific documents that are retrieved from a dataset. Amazon Q has a built-in RAG system. A RAG model has the following two components:

• A retrieval component retrieves relevant documents for the user query.

• A generation component takes the query and the retrieved documents and then generates an answer to the query using a large language model.

Large language model

A large language model (LLM) is a language-based, machine learning model that's tuned to a large number (billions) of parameters or and trained on a large corpus of documents.

Retriever

A retriever pulls data from an index in real time during a conversation. Amazon Q supports a native index retriever and also a Amazon Kendra index retriever.

Index

An index is a corpus of documents. Amazon Q supports its own index where you can add and sync documents. An index has fields that you can map your document attributes to enhance your end user's chat experience. Amazon Q creates an index for you when it creates your Amazon Q native retriever. If you use the API, Amazon Q Business provides two types of index: Enterprise and Starter. If you use the console, Amazon Q Business creates an Enterprise index by default.

You can also use an Amazon Kendra index as a retriever for your generative AI application.

Data source

A data source is a document repository.

Data source connector

A data source connector can crawl and synchronize a data source with an Amazon Q index at customizable intervals. Amazon Q supports multiple connectors so that you can build your generative AI solution with minimal configuring. For a list of Amazon Q supported connectors, see Supported connectors. For an overview of Amazon Q connector features, see Amazon Q data source connector features.

IAM Identity Center

You can manage user access to your Amazon Q Business application using IAM Identity Center as your AWS gateway to the identity provider of your choice. For more information on creating an Amazon Q Business application integrated with IAM Identity Center see Configuring an Amazon Q Business application. For more information about using IAM Identity Center to manage access to applications, see Manage access to applications in the IAM Identity Center User Guide.

Document

In Amazon Q, a document is a unit of data. Specific document formats supported include .csv, .docx, HTML, JSON, .pdf, plaintext, .ppt, .pptx, .rtf, and .xslx. For more information, see Supported document types.

Application

An Amazon Q application is the primary resource that you use to create a chat solution. To create the application, you can use either the Amazon Q console or Amazon Q API actions.

Web experience

An Amazon Q web experience is the chat interface that you create using your Amazon Q application. Then, your end users can chat with your organization’s Amazon Q web experience. You can configure and customize your Amazon Q web experience using either the Amazon Q console or the Amazon Q API.

Guardrails and chat controls

An Amazon Q feature that lets you define global controls and topic-level controls for your application. Using this feature, you can control what sources your application will use to generate responses from, and also control what topics it will respond to and how. For more information, see Guardrails.

Plugins

Amazon Q includes a plugins feature that you can use to interact with third-party services such as Jira and Salesforce. With the plugins feature, you can perform actions specific to that service (like creating a ticket) from within your Amazon Q web experience chat. For more information, see Plugins.

Quick prompts

The Amazon Q quick prompts feature helps with end user discoverability of the web experience chat features. Use this feature to prompt your end user to engage with their web experience chat in specific ways. For example, you can show the available configured plugins or inform users that they can choose to summarize their chat.

Document attributes

Document attributes are structural metadata associated with documents, such as document title, document type, and date and time created. Amazon Q extracts document attributes during the document ingestion process to provide customizable chat and data manipulation capabilities for your application. Amazon Q offers reserved document attributes that you can use. Or, you can create custom attributes. For more information, see Document attributes, Filtering using document attributes, Boosting using document attributes, and Custom document enrichment.

Filtering using document attributes

Filtering using document attributes is an Amazon Q feature that you can use to filter your Amazon Q chat responses for your end user. For example, if you have a document attribute associated with a data source type, you can use the attribute to mandate that chat responses only be generated from a specific data source. For more information, see Filtering using document attributes.

Relevance tuning

You can choose to use document attributes to boost and tune the relevance of chat responses for end users from specific content. For example, if you have a document attribute associated document creation or updation date, you use these attributes to boost chat responses from more recently created or updated documents. For more information, see Relevance tuning.

Document enrichment

Document enrichment is an Amazon Q feature that you can use to manipulate your document content and document attributes. You can use document enrichment to perform optical character recognition (OCR) or translation. Document enrichment uses basic and Lambda operations. For more information see, Document attributes and types and Document enrichment.

Field mappings

An Amazon Q index has fields that help you structure data to aid the retrieval process. You can map index fields to your document attributes when you add documents directly to an index, or use a data source connector.

User Store

User Store is an Amazon Q data source connector feature that streamlines user and group management across all the data sources attached to your application. For more information about how this feature works and implementation details, see Understanding User Store.

Index capacity

When you use an Amazon Q native retriever for your application, you must provision data storage capacity for your index. If you use the API, Amazon Q Business provides two types of index: Enterprise and Starter. If you use the console, Amazon Q Business creates an Enterprise index by default. Both index types include 20,000 documents or 200 MB of total extracted text (whichever is reached first) and 100 hours of data connector usage (time that it takes to scan and index new, updated, or deleted documents). For more information, see Amazon Q Index types and Pricing for subscriptions and indices.

Tags

Manage your Amazon Q applications and data sources by assigning tags or labels. You can use tags to categorize your Amazon Q resources in various ways. For example, categorize by purpose, owner, or application, or any combination. Each tag consists of a key and a value, both of which you define. For more information, see Tags.

Foundation model

A foundation model (FM) is a broad, function-based machine learning model (not specific to language systems). An FM is tuned to a large number (billions) of parameters and is trained on a large corpus of documents.

Hallucination

A hallucination, in the machine learning context, is a confident response by an AI application that isn't justified by its training data. Think of a hallucination as instances where the response doesn't make sense in the context of the prompt, or when the responses are out of scope with the documents provided. Amazon Q offers you the ability to minimize hallucinations by allowing your retrieval system to generate responses only from your existing enterprise data.