Glossary / Terminology Explained at High Level
Glossary / Terminology Explained at High Level
Amazon Bedrock:
Amazon Bedrock is a fully managed service that enables developers to build and scale generative AI applications using foundation models (FMs) from leading AI companies and Amazon. It provides a unified API to access these models, allowing developers to experiment, customize, and integrate them into their applications without managing infrastructure.
Foundation Model (FM):
A foundation model is a large, pre-trained machine learning model capable of performing various tasks, such as text generation , summarization , language translation , and more. Amazon Bedrock offers access to foundation models like GPT-4 and PaLM from multiple providers, enabling developers to create advanced AI applications without starting from scratch.
https://www.skool.com/k21academy/classroom/f1bc0b14?md=edf7f1615edb4a3e8fd7eeb25551d62f
Model Provider:
A model provider is the entity that develops and hosts a foundation model. For example, Amazon Bedrock allows access to models from providers such as Anthropic , Meta , and AI21 Labs. These models can be accessed via a unified API in Amazon Bedrock for use in applications hosted on AWS Cloud.
Amazon Q Developer:
Amazon Q Developer is a Generative AI-powered assistant specifically designed for developers. It helps automate tasks like code generation, debugging, and understanding through natural language, enabling developers to save time and improve productivity. This service integrates with various AWS services , such as AWS Lambda and AWS CloudFormation , to enhance the development workflow.
Inference:
Inference is the process where a trained AI model takes an input (prompt) and generates an output. This happens after the model has been trained on large datasets. Amazon SageMaker is a popular AWS service used for inference, where models can be deployed and used for real-time or batch predictions.
Asynchronous Inference:
Asynchronous inference allows you to submit a request to an AI model, process it in the background, and receive results at a later time. This is useful for tasks that require more processing power or involve large datasets, and is often used in AWS AI services like Amazon SageMaker or AWS Lambda , where the response time is not crucial to the workflow.Synchronous Inference:
Synchronous inference is when a model immediately generates a response after receiving a prompt. This is ideal for real-time applications like chatbots or data analysis tools. In AWS , SageMaker or AWS Lambda is often used for synchronous inference, ensuring fast and responsive interactions.Batch Inference:
Batch inference processes multiple data inputs at once. It is often used for tasks like processing a large set of data, running predictions, or analyzing datasets that don't require immediate feedback. In AWS , this is commonly done with Amazon SageMaker or AWS Batch for high-volume AI/ML tasks.
Guardrails:
Guardrails in Amazon Bedrock are security and content moderation tools that prevent the model from generating harmful or inappropriate content. These can be set up to filter specific types of input or output (e.g., hate speech or offensive language), ensuring that your AI applications align with ethical standards.
Prompt:
A prompt is the input provided to a foundation model to guide its output. For example, you might prompt a model with “Translate this text into Spanish” or provide an example dialogue. Amazon Bedrock allows developers to fine-tune prompts to improve the relevance and creativity of the model’s responses.
- Prompt Engineering:
Prompt engineering is the process of designing and refining prompts to generate more accurate, relevant, and creative responses from AI models. In AWS, developers use Amazon Bedrock to experiment with prompt designs, optimizing the input for desired outcomes, such as specific tone or style in text generation.
Few-Shot Prompting:
Few-shot prompting involves providing a foundation model with a few examples to guide its behavior. For example, if you want the model to translate sentences in a specific style, you can provide a few examples of that style. This method is particularly useful in AWS Bedrock , where models can be fine-tuned with minimal training data to perform specific tasks.Zero-Shot Prompting:
Zero-shot prompting refers to giving a foundation model a task without providing any examples. The model must use its pre-trained knowledge to perform the task. This method is widely used in Amazon Bedrock to handle general tasks like text translation or summarization.
Max Tokens:
Max tokens refer to the limit set for how many tokens (units of text) a model can generate in a response. In Amazon Bedrock , the max tokens setting helps control the length of the model’s response, which is important for both performance and cost when interacting with large models like GPT-4.
SDK (Software Development Kit):
A SDK is a set of tools and libraries that allows developers to build and integrate applications with AWS services. For example, AWS SDK for Python (Boto3) can be used to integrate Amazon Bedrock APIs into your applications, enabling developers to manage resources like SageMaker models or Lambda functions directly from the code.
Temperature:
Temperature is a setting that controls the randomness of the model’s output. A higher temperature leads to more varied and creative responses, while a lower temperature results in more focused and consistent output. This setting is often used in Amazon Bedrock to influence the tone and creativity of text generation.
Token:
A token is a unit of text (such as a word or a part of a word) that the model processes. In AWS services like SageMaker and Amazon Bedrock , both the input prompts and model responses are measured in tokens, which helps manage costs and performance when using AI models.
Unified API:
Amazon Bedrock provides a unified API that allows developers to interact with different foundation models consistently, simplifying integration across multiple AI models in your applications hosted on AWS Cloud.
Action Group:
An action group is a collection of functions, like AWS Lambda , that a Bedrock Agent can trigger to complete tasks. These can be actions like sending an email, processing data, or invoking another AWS service within a workflow.
Agent:
An agent is a managed orchestration system in Amazon Bedrock that handles multi-step tasks using foundation models, tools, and memory. It can automate workflows, making it ideal for applications that require sequential tasks based on user input.
Amazon Bedrock Flows:
Amazon Bedrock Flows is a no-code tool that allows you to chain together different tasks, like prompting models or invoking AWS tools, into an automated workflow. This is useful for building complex AI applications without writing a lot of code.
Amazon Bedrock Knowledge Bases:
This feature allows you to connect foundation models to private data stored in Amazon S3 , enabling retrieval-augmented generation (RAG). You can use your own data to enhance the accuracy and relevance of the model's responses.
Chunking:
Chunking refers to the process of breaking large documents into smaller pieces (chunks). This helps models process the content more efficiently during inference, especially in Amazon Bedrock when retrieving information from a knowledge base.
Condition Node:
A condition node in Bedrock Flows allows you to add logic that decides which path to take based on input values or model outputs, creating dynamic workflows that change based on real-time data.
Converse API:
Amazon Bedrock's Converse API allows developers to build conversational applications by managing session context and orchestrating multi-turn interactions between users and foundation models.
Embeddings:
Embeddings are numerical representations of text that help models compare and understand the semantic meaning of different pieces of information. In Amazon Bedrock , embeddings allow models to search for similar content in knowledge bases during inference.
Iterator Node:
An iterator node in Bedrock Flows lets you loop over a list of items and apply logic to each item, making it easier to handle repeated tasks in your workflows, such as processing a batch of data.
Knowledge Retrieval:
Knowledge retrieval involves searching for and using relevant data from an external source (like an Amazon S3 bucket) to enhance the response generated by the model.
Memory:
Memory in Amazon Bedrock Agents allows models to store and recall context from previous interactions, enabling more personalized and context-aware conversations across multiple sessions.
Prompt Management:
Prompt management allows developers to store, reuse, and version their prompt templates. This ensures consistency in responses and saves time when using prompts for multiple applications in Amazon Bedrock.
RetrieveAndGenerate:
A process where the model first retrieves relevant information from a knowledge base and then uses that information to generate a more accurate and context-rich response.
Semantic Chunking:
Semantic chunking divides text based on its meaning rather than fixed size. This improves the relevance of information retrieved from knowledge bases during tasks like document summarization or question-answering.
Tool:
A tool is an external function or API, like AWS Lambda , that a model can call to perform additional tasks. For example, a model might use a Lambda function to calculate a result or fetch data from an external source.
SageMaker – A fully managed AWS service for building, training, and deploying machine-learning models.
Lambda – A serverless compute service that runs your code in response to events without managing servers—often used as tools by AI Agents.
CloudFormation – An AWS service that lets you create and manage resources through code templates (infrastructure as code).
API Gateway – Used to expose AI or backend services as REST APIs for external applications.
IAM (Identity and Access Management) – Controls who can access which AWS resources; crucial for AI and data security.
Elastic Inference – An AWS feature that lets you attach just the right amount of GPU acceleration to a SageMaker model for faster predictions.
Vector Database – A special database that stores embeddings and enables semantic search in AI apps (used in RAG systems with Bedrock).
Data Lake – A centralized repository (e.g., Amazon S3) that stores structured and unstructured data for analytics and AI training.
CloudWatch – An AWS monitoring tool that tracks usage, performance, and costs of your AI/ML resources.
API Key – A secure identifier used to authenticate requests when calling Bedrock or other AWS services via code.
Endpoint – The URL through which your AI model or API is accessed for real-time predictions or responses.
“Semantic”
“Semantic” = Understanding the meaning instead of just reading the words.
It’s what allows AI systems like Amazon Bedrock , OpenAI , and SageMaker models to respond intelligently and contextually.
Semantic comes from the word semantics , which means “related to meaning” — not just the words themselves, but what those words mean in context.
In AI and cloud computing , semantic refers to understanding or processing information based on meaning and context , rather than just matching text or data literally.