AI Terms Explained
Your guide to AI terminology explained in simple terms by Alltegrio AI development experts.
A
Agent Memory
Agent memory is the stored information an AI agent can read back later, either within one conversation (short-term) or across conversations and users (long-term), kept outside the model itself.
Agent Orchestration
Agent orchestration is the control layer that decides which model call, tool or agent runs next, keeps the shared state, saves progress, and pauses for people when a step needs approval.
Agent2Agent Protocol (A2A)
The Agent2Agent Protocol (A2A) is an open standard that lets AI agents built by different teams or vendors discover each other, delegate tasks and exchange results over HTTP without exposing their internal logic.
Agentic AI
Agentic AI describes software in which a language model works toward a goal over several steps. It plans, calls tools and APIs, reads the results and decides the next step, inside limits that people set.
AI Agent
An AI agent is a program in which a language model chooses and calls tools in a loop, reads each result, and decides the next step until a stop condition ends the run.
AI Governance
AI governance is the set of policies, roles, processes and technical controls an organisation uses to decide which AI systems it builds or buys, and to keep them safe, lawful and accountable while they run.
AI Guardrails
AI guardrails are the checks and limits placed around a language model application, on its inputs, its outputs and the actions it can take, so it stays within its intended scope, follows policy and cannot cause damage when the model gets something wrong.
AI Hallucination
An AI hallucination is output from a language model that sounds confident and plausible but is false or unsupported, such as an invented fact, quote, source, figure or API.
AI Observability
AI observability is the practice of recording what an AI application did on each request, including prompts, retrieved data, tool calls, outputs, tokens, cost and latency, so teams can explain, debug and improve its behaviour.
Ambiguity
Ambiguity is when a word, sentence, request or piece of data can reasonably be read in more than one way. AI systems have to resolve it too, and a wrong guess produces a confident answer about the wrong thing.
Annotation
Annotation means adding comments, explanatory notes, or labels to any information. The practice has found broad applications in fields such as literature, linguistics, and computer science as it provides context or…
C
Chatbot
A chatbot is a program that holds a conversation with a person in text or voice. Most chatbots built today use a large language model to understand the request and look up facts in company data instead of answering from memory.
Computer Vision
Computer vision is the branch of artificial intelligence (AI) that lets computers extract useful information from images and video. Computer vision systems recognize objects, detect defects, read text in pictures, and track motion.
Computer-Use Agent
A computer-use agent is an AI agent that operates software the way a person does, by looking at screenshots and sending mouse clicks and keystrokes, so it can work with applications that have no API.
Context Engineering
Context engineering is the practice of deciding what information a language model sees at each step (instructions, retrieved data, tool definitions, tool results and conversation history) so it has what it needs and nothing that distracts it.
Context Window
A context window is the maximum amount of text, measured in tokens, that a language model can take into account in one request, including the instructions, the documents and history you send, and the answer it writes.
D
Data Augmentation
Data augmentation creates extra training examples by applying realistic changes to existing data, such as cropping an image, rephrasing a sentence or adding noise to audio, while keeping each example's label true.
Data Monetization
Data monetization is the process of transforming data into valuable assets that generate economic benefit. Companies use their information to improve operations and customer experience, and to create new sources of revenue.
Data Residency
Data residency is the requirement or guarantee that data is stored, and often processed, in a specific country or region. For AI systems it covers prompts, documents, embeddings, logs and model outputs, not only the database.
Deep Learning
Deep learning is a branch of machine learning that trains neural networks with many layers to learn patterns directly from raw data such as images, text and audio. Almost every modern AI system, including large language models, is built with it.
Deterministic
Deterministic can be defined as a system, process, or algorithm that provides complete predictability and uniformity in results, provided the input conditions are the same. To put it simply, a process is deterministic if it repeatedly yields the same output. Deterministic systems come in handy while working with AI because they add both reliability and transparency to any process.
Discriminative AI
A discriminative model learns to tell categories apart, or to predict a value, from the features of an input. Unlike a generative model, it does not learn how the data itself is distributed, so it classifies and scores rather than producing new examples.
E
Embeddings
Embeddings are lists of numbers (vectors) that a model produces to represent the meaning of text, images or other data, so that similar items get vectors that sit close together and can be compared mathematically.
EU AI Act
The EU AI Act (Regulation (EU) 2024/1689) is the European Union's law on artificial intelligence. It sorts AI systems by risk, bans some practices, sets duties for high-risk systems and general-purpose AI models, and requires transparency.
Extensibility
Extensibility is the capability of a system, software, or framework to expand, evolve, or be extended easily without major changes. Think of it as flexibility in a system that takes on new challenges or expands its…
F
Fine-Tuning
Fine-tuning is further training of an already trained model on a smaller, task-specific set of examples, so it adopts a particular format, style, behaviour or skill more reliably than prompting alone can achieve.
Foundation Model
A foundation model is a large AI model trained on broad data at scale that can be adapted to many different tasks. Large language models, image generators and speech models are all foundation models.
Function Calling (Tool Use)
Function calling, also called tool use, lets a language model ask the application to run a defined function, such as a database query or API call, by returning the function name and arguments. The application runs it and returns the result.
L
Large Language Model (LLM)
A large language model (LLM) is a neural network trained on very large amounts of text to predict the next token. That one skill lets it answer questions, write, summarise, translate, generate code and call tools.
LLM Evaluation (Evals)
LLM evaluation, usually called evals, is the practice of testing a language model application against a fixed set of realistic tasks with defined success criteria, so a team can measure quality and catch regressions before and after every change.
LLMOps
LLMOps (large language model operations) is the set of practices and tools for shipping and running LLM applications in production, covering prompt and model versioning, evaluation, monitoring, cost control and safety.
M
Model Context Protocol (MCP)
The Model Context Protocol (MCP) is an open standard that defines how AI applications connect to external tools and data sources, so one integration (an MCP server) can be used by any AI application that supports the protocol.
Model Distillation
Model distillation trains a small student model to imitate a larger teacher model on the tasks you care about. The student keeps much of the teacher's quality at a fraction of the cost and latency.
Model Routing
Model routing is sending each request to the language model best suited to it, usually a cheaper, faster model for easy requests and a stronger one for hard ones, based on rules, a classifier or a fallback chain.
Multi-Agent System
A multi-agent system is an AI application in which several agents, each with its own instructions and tools, split a task between them and pass results to one another or to a coordinating agent.
Multi-Modal AI
Multi-modal AI takes in, and often produces, more than one type of data, such as text, images, audio and video, within a single model. A text-only language model reads and writes text; a multi-modal model can use a photo, a voice note and a question together.
N
Natural Language Generation
Natural language generation (NLG) is the part of AI that produces readable text from an input such as a table of figures, a database record, a document or a typed request. Today most NLG in products is done by large language models.
Neural Network
A neural network is a machine learning model made of layers of simple connected units whose connection weights are learned from data. It is loosely inspired by the brain and is the basis of deep learning and modern AI.
P
Probability Model
A probabilistic model, also called a probability model, is a mathematical model that predicts outcomes while accounting for random events. It assigns a probability to each possible outcome instead of giving a single forecast.
Prompt Caching
Prompt caching is a feature of LLM APIs that stores the processed beginning of a prompt so later requests starting with the same text are billed at a lower rate and start answering faster.
Prompt Injection
Prompt injection is an attack in which text supplied to a language model, typed by a user or hidden in a document, email or web page the model reads, changes the model's behaviour against the intentions of the application's owner.
Prompting
In AI, prompting is the practice of writing the input that tells a generative model what to do, including the instructions, the context and examples it should use, and the shape of the answer you expect.
R
Reasoning Model
A reasoning model is a large language model trained to work through a problem step by step before it answers. It spends extra tokens and time on thinking to get better results on maths, code, planning and multi-step analysis.
Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation (RAG) is a pattern in which a system first finds relevant passages in your own documents or data and then asks a language model to answer using only that material, with references to the sources.
S
Small Language Model (SLM)
A small language model (SLM) is a language model compact enough to run cheaply on a single GPU, a laptop or a phone. It trades broad knowledge for speed, low cost and the option to keep data on your own hardware.
Stacking
Stacking, or stacked generalization, is an ensemble technique in machine learning in which several models make predictions and a separate meta-model learns how best to combine them into a final answer.
Structured Data
Structured data is information that is neatly organized into a standardized format. This data format allows performing search, processing, and analysis easily. Think of a spreadsheet where every piece of information…
Structured Outputs
Structured outputs are model responses constrained to a schema the developer supplies, usually JSON Schema, so the application receives fields of known names and types instead of free text it has to parse.
Summarization
Summarization involves condensing a lot of text into a smaller version but retaining the key points and the general meaning. In the world of artificial intelligence, summarization is one of the most valuable features of…
Synthetic Data
Synthetic data is data generated by an algorithm or model instead of collected from real events. It is used to test systems, fill gaps in training data and protect privacy, but it only helps when checked against real data.
T
Technical Due Diligence
Technical due diligence is a complex investigation into the quality, risks, and scalability of a target company’s technology assets, processes, and infrastructure. It is frequently necessary in mergers and acquisitions…
Text-to-SQL
Text-to-SQL is the use of a language model to turn a question in plain language into an SQL query, run it against a database and return the result. It lets people ask data questions without writing SQL themselves.
U
Unstructured Data
Large amounts of data are generated every second in our modern digital world. While some of the data is in tabular or database format, most of the data exists freely in an unorganized manner. Data of this nature is referred to as unstructured data.
Unsupervised Learning
Unsupervised learning is a type of machine learning in which an algorithm finds structure in unlabeled data, such as groups of similar items, unusual items or a simpler representation, while people decide what the patterns mean.
V
Vector Database
A vector database stores embeddings, the numeric vectors that represent text, images or other data, and finds the stored items whose vectors are closest to a query vector. It is the search layer behind most semantic search and RAG systems.
Voice Agent
A voice agent is an AI system that holds a spoken conversation, by phone or in an app, and can act on it: look up an order, book an appointment or route a call. It combines speech recognition, a language model, tools and speech synthesis.