AI Concepts: Core Components of Modern AI

Modern AI systems are built from interconnected components. Large language models generate responses. Embeddings convert data into numerical representations. Vector databases store and search those representations at scale. And retrieval systems connect everything to real-world information. Understanding how these pieces fit together is essential for anyone building, deploying, or making decisions about AI.

Core Concepts

00 · Foundation

Transformers

The neural network architecture built on attention that underpins nearly all modern AI. Introduced in 2017, it is the design beneath every large language model, and the parent of this whole map.

01 · Parent Concept

Large Language Models (LLMs)

Neural networks trained on massive text datasets that can understand, generate, and reason about human language. The technology behind ChatGPT, Claude, and Gemini.

02 · Connector

Embeddings

Numerical representations that capture meaning in mathematical form, enabling AI to search, compare, and reason about information by semantic similarity.

03 · Infrastructure

Vector Databases

Specialised data systems that store and search high-dimensional vectors, enabling AI applications to find relevant information in under 50 milliseconds.

04 · Application

Retrieval-Augmented Generation (RAG)

The architecture that grounds language model responses in current, verified information by combining retrieval with generation.

05 · Autonomy

AI Agents

Autonomous software systems that combine LLM reasoning with tool execution, memory, and feedback loops to complete multi-step tasks independently.

06 · Coordination

Multi-Agent Systems (MAS)

AI architectures where multiple autonomous agents collaborate and coordinate to accomplish complex tasks. The organizational layer powering enterprise agentic AI in 2026.

07 · Control

AI Agent Orchestration

The control layer that decomposes goals, routes tasks to agents and tools, arbitrates memory, and assembles auditable results. How autonomous work stays coordinated.

08 · Reliability

AI Hallucinations

Fluent, confident outputs that are factually wrong or ungrounded. Why generative models make things up, and how grounding, verification, and abstention reduce it.

09 · Execution

Tool Use & Function Calling

The interface that lets a model call external software through function calling, receive the results, and act on the world. The bridge between reasoning and action beneath every AI agent.

How these concepts connect

Transformers are the underlying architecture, large language models generate responses, embeddings convert data into vectors, vector databases store and retrieve them, retrieval systems connect everything to real-world information, tool use lets models act on that information, and AI agents orchestrate these components to take autonomous action. Together they form the standard architecture for enterprise AI deployment in 2026.

Transformers

→

LLMs

→

Embeddings

→

Vector DBs

→

RAG

→

Tool Use

→