iGentixAI Periodic Table of AI Elements

A visually engaging framework for understanding the world of AI terms
Created by a Generative Learning class • For educational use

www.igentix.ai HOME

The iGentixAI Periodic Table is a conceptual framework that organizes the rapidly evolving field of AI into a familiar structure, similar to the chemical periodic table. It categorizes AI concepts into families (columns) representing functional roles and periods (rows) representing levels of complexity.

This table helps you organize chaotic AI terminology, decode real-world AI architectures and product demos, and predict how different "elements" combine to form complete systems (e.g., a company documentation chatbot using Embeddings → Vector Databases → RAG → Prompt → LLM → Guardrails, or an agentic loop using Agents → Function Calling → Frameworks).

Click any element below to jump to its detailed properties and glossary entry.

Reactive
Retrieval
Orchestration
Validation
Models
Primitives
PRPrompt
EMEmbeddings
LLMLarge Language Model
Compositions
FCFunction Calling
VXVector Databases
RAGRetrieval Augmented Generation
GRGuardrails
MMulti-modal Models
Deployment
AGAgents
FTFine-tuning
FWFrameworks
RTRed Teaming
SMSmall Models
Emerging
MAMulti-agent Systems
SYSynthetic Data
INInterpretability
THThinking Models

Glossary & Properties

Prompt (PR) — Reactive Family, Primitives Period

Definition: Instructions given to an AI (e.g., “write me an email” or “summarize this document”).

Properties & Elaboration: The most basic way to interact with AI models. Effective prompting is an art (prompt engineering) using techniques like zero-shot (no examples), one-shot/few-shot (provide examples), and chain-of-thought (encourage step-by-step reasoning) to achieve higher-quality, more reliable outputs.

Embeddings (EM) — Retrieval Family, Primitives Period

Definition: Numerical vector representations of meaning (typically ~1536 dimensions) that capture semantic similarity.

Properties & Elaboration: Text is transformed into vectors where similar meanings are mathematically close (e.g., “vacation policy” and “time off guidelines” have nearby vectors). This enables semantic search rather than keyword matching, foundational for modern retrieval systems.

Large Language Model (LLM) — Models Family, Primitives Period

Definition: Foundational models (e.g., ChatGPT, Claude, Gemini, Grok) trained on massive datasets to understand and generate human-like text.

Properties & Elaboration: Built on transformer architecture with large context windows (short-term memory measured in tokens). They are the stable “noble gas” core that everything else reacts with or builds upon. Context window size varies widely and affects practical performance on long documents.

Function Calling (FC) — Reactive Family, Compositions Period

Definition: Ability of an LLM to invoke external tools/APIs before answering (e.g., calling a weather API).

Properties & Elaboration: Enables real-world interaction and dynamic data retrieval, a key building block for agentic behavior.

Vector Databases (VX) — Retrieval Family, Compositions Period

Definition: Specialized databases (e.g., Pinecone, ChromaDB) optimized for storing and querying embeddings via semantic similarity.

Properties & Elaboration: Support fast approximate nearest-neighbor search at scale. Key configuration includes dimensionality, similarity scoring thresholds, and chunk overlap to preserve context when splitting large documents.

Retrieval Augmented Generation (RAG) — Orchestration Family, Compositions Period

Definition: Process of retrieving relevant context from external sources and injecting it into the prompt before generation.

Properties & Elaboration: Combines semantic search (embeddings + vector DB) with generation to provide up-to-date, factual, domain-specific answers while reducing hallucinations. Classic pattern: query → embedding → retrieve relevant chunks → augment prompt → generate.

Guardrails (GR) — Validation Family, Compositions Period

Definition: Runtime safety filters and schema validation to ensure appropriate, safe outputs.

Properties & Elaboration: Prevent harmful content, enforce response formats, and protect against prompt injection or undesirable behavior.

Multi-modal Models (M) — Models Family, Compositions Period

Definition: LLMs capable of processing multiple data types (text, images, audio, video).

Properties & Elaboration: Extend text-only LLMs to understand and reason over visual or auditory inputs (e.g., GPT-4o, Gemini).

Agents (AG) — Reactive Family, Deployment Period

Definition: Autonomous AI systems using think-act-observe loops to achieve goals.

Properties & Elaboration: Unlike static LLMs, agents plan, use tools (via function calling), observe results, and iterate. They possess autonomy, memory, and tool access, enabling complex real-world task completion.

Fine-tuning (FT) — Retrieval Family, Deployment Period

Definition: Adapting a base model by further training on domain-specific data.

Properties & Elaboration: Improves performance on specialized tasks (e.g., medical or legal domains) by customizing model weights.

Frameworks (FW) — Orchestration Family, Deployment Period

Definition: Platforms/libraries (e.g., LangChain, LangGraph) that integrate components for building and deploying AI systems.

Properties & Elaboration: Provide memory management, tool integration, orchestration of multi-step workflows, and easy switching between LLM providers. LangGraph excels at complex stateful agent workflows.

Red Teaming (RT) — Validation Family, Deployment Period

Definition: Adversarial testing to expose vulnerabilities (jailbreaks, prompt injection, data exfiltration).

Properties & Elaboration: Essential for identifying and hardening safety weaknesses before production deployment.

Small Models (SM) — Models Family, Deployment Period

Definition: Distilled/specialized models that are fast, cheap, and can run on-device.

Properties & Elaboration: Trade some capability for efficiency and low latency, suitable for edge deployment.

Multi-agent Systems (MA) — Reactive Family, Emerging Period

Definition: Multiple specialized agents collaborating, debating, and coordinating to solve complex problems.

Properties & Elaboration: Enable division of labor and emergent capabilities beyond single-agent systems.

Synthetic Data (SY) — Retrieval Family, Emerging Period

Definition: AI-generated training data used when real data is scarce or sensitive.

Properties & Elaboration: Helps bootstrap or augment training datasets, especially for specialized domains.

Interpretability (IN) — Validation Family, Emerging Period

Definition: Techniques to understand why a model makes specific decisions.

Properties & Elaboration: Aims to open the “black box” by identifying influential components and reasoning paths.

Thinking Models (TH) — Models Family, Emerging Period

Definition: Models designed to reason internally (often via built-in chain-of-thought) before answering.

Properties & Elaboration: Deliberately spend compute on planning and reflection for higher-quality, more reliable outputs on complex tasks.