The Technovids learning library
AI & Cloud Glossary
AI terms explained: understand the term, see it in action, and know what to learn next.
Explore AI, data and cloud concepts through plain-language explanations, practical examples and useful diagrams. Start with a learning path or jump straight to a topic.
Not sure where to begin?
Choose a starting path
Use AI at work
Understand what AI produces, then learn to guide and check it.
Build an AI application
Move from models and retrieval to tools and agent behaviour.
Connect data and cloud
Follow the foundations behind analytical applications.
Explore the concepts
AI foundations
Machine Learning (ML)
Machine learning is a branch of AI in which a model learns patterns from data to perform a task, rather than relying only on explicitly written rules. Its usefulness depends on how well those patterns work on relevant new examples, not just the training data.
Neural Network
A neural network is a machine-learning model made of connected units that transform inputs using settings it has learned. Layers build on each other to recognise patterns. Training adjusts the settings; using the model applies them to produce an answer.
Foundation Model
A foundation model is trained on broad data at scale and can be adapted to a range of downstream tasks. It provides a reusable starting point, not a finished application or a guarantee that it will perform well in a particular organisation.
Generative AI
Generative AI is a kind of AI that creates content such as text, images, code or audio, based on patterns it has learned and the input you give it. What it produces is a draft answer, not automatically a fact it looked up, an original idea or a checked result.
Large Language Model (LLM)
A large language model (LLM) is a machine-learning model trained on large amounts of language data to learn patterns that support tasks such as text generation. Many contemporary LLMs generate responses by predicting successive tokens from the input and the text already produced.
Natural Language Processing (NLP)
Natural language processing (NLP) is the area of computing that analyses, understands and produces human language. It covers tasks like sorting text, pulling out details, translating and summarising, using methods from simple rules to large language models.
Transformer Architecture
A transformer is a type of neural network that uses attention to combine information from different positions in a sequence. Attention is not the whole model: transformers also use feed-forward layers and position information.
Multimodal AI
Multimodal AI works with more than one type of information, such as text and images or audio and video. A multimodal application can relate a question to visual or other non-text input; its supported input and output types depend on the model and system design.
Explore the concepts
Building AI applications
Prompt Engineering
Prompt engineering is the practice of designing and testing the instructions, context and examples supplied to an AI model to achieve a defined task. It involves evaluating outputs and refining the input, rather than relying on a particular phrase to guarantee a good response.
Embeddings
Embeddings are a way of turning data into numbers, usually lists of numbers called vectors. An embedding model learns to place related items close together, so you can measure how similar they are. Apps use them for things like meaning-based search, recommendations and grouping similar content.
Vector Database
A vector database stores and searches vectors, such as embeddings. It helps apps find similar items and keep track of the records they belong to. Vector search can come from a specialist database, or be added to a database you already use.
Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation (RAG) is a method that finds relevant information and gives it to a generative model when it answers a question. It links the answer to outside sources, such as product documents or company policies, without retraining the model each time those sources change.
Fine-Tuning
Fine-tuning is extra training that adapts a pretrained model using a chosen set of examples. It changes some or all of the model’s settings so it does better at a particular task or subject. For language models, the examples might teach a consistent answer format, a way of sorting requests or a specific behaviour.
AI Agent
An AI agent is a system that selects actions towards a goal, observes what happens and uses that feedback to decide what to do next. In an LLM-based agent, a language model helps choose actions while application code manages tools, permissions and execution.
Agentic AI
Agentic AI describes an approach in which AI systems choose and carry out steps towards a goal, adjusting their actions from feedback. The term describes how a system behaves; an AI agent is a component or system that exhibits that behaviour.
Tool Calling
Tool calling is a mechanism through which a model requests a named operation with structured arguments, such as an order lookup. A tool runtime executes the operation and returns its result; generating a tool request does not itself execute the action.
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open protocol for connecting AI applications to external capabilities and context. An MCP host manages clients that connect to servers exposing supported tools, resources or prompts.
LangChain
LangChain is an open-source framework for building applications and agents around language models. It provides model and tool integrations and an agent harness; it is not itself a language model or a source of business data.
Explore the concepts
Quality and responsible use
AI Hallucination
An AI hallucination is generated content that is false or unsupported while being presented as an answer. In a document-based task, this can include details not supported by the supplied source, even when those details sound plausible. Fluent wording and confident tone do not establish accuracy.
AI Bias
AI bias is a systematic tendency in an AI system’s data, design or use that can produce skewed results or unfair effects. Evaluating bias requires attention to the affected people and context, not just a single overall accuracy score.
Explainable AI (XAI)
Explainable AI refers to methods that help people understand a model’s behaviour or a particular output. An explanation should be appropriate to its audience and faithful to what it describes; a plausible story alone is not evidence of how a model reached its result.
Responsible AI
Responsible AI means thinking about how AI affects people at every stage, from design to everyday use, and acting to reduce harm you can foresee. It links goals such as reliability, fairness, privacy and accountability to real decisions and evidence.
AI Governance
AI governance is the set of responsibilities, policies and decision processes through which an organisation oversees AI across its lifecycle. It establishes who can approve a use, what evidence is required and who responds when the system causes problems.
Explore the concepts
Data and analytics
Data Pipeline
A data pipeline is a sequence of processes that moves data from sources to destinations, often validating or transforming it along the way. A reliable pipeline manages failures, duplicates and freshness as well as transporting records.
Data Warehouse
A data warehouse is a system built for reporting and analysis across data brought together from many sources, often with years of history. It is only useful if the data has consistent definitions and a clear structure, not just because everything sits in one database.
Business Intelligence (BI)
Business intelligence (BI) is the practice of organising, analysing and presenting business data to help people make decisions. It combines agreed measures, data models and reports. A good-looking dashboard does not prove its numbers mean anything.
MLOps
MLOps is the practice of building, releasing and running machine-learning systems with repeatable steps and monitoring. It links code, data, model versions and testing, so a live prediction service can be looked after and investigated.
Explore the concepts
Cloud and delivery
Cloud Computing
Cloud computing gives you on-demand access, over the network, to shared computing resources that you can set up and release with little manual work. It changes how you get and run resources; you still have to manage your apps, access, data and costs.
Infrastructure as a Service (IaaS)
Infrastructure as a Service (IaaS) is a cloud model that gives you computing power, storage and networking, while you look after the operating system and the apps on it. Exactly who looks after what depends on the provider and the service.
Serverless Computing
Serverless computing is a cloud model where the provider sets up the servers and runs most of the capacity for you. Developers set how the app behaves and its limits, instead of looking after individual servers.
Cloud Native
Cloud native is a way of designing and running software so it can change often, recover from failure and be monitored, in computing environments that keep changing. It is a set of design and running practices, not just the fact that an app is hosted in the cloud.
Docker
Docker is a set of tools for building, sharing and running apps in containers. An image packages the app’s files and everything it needs; a container is a running copy of that image, with its own settings and isolation.
Kubernetes (K8s)
Kubernetes is an open-source platform for running and managing containers. It keeps checking what is actually running and works to match what you asked for, such as keeping a set number of copies of an app running.
DevOps
DevOps is a way of working that links software development and day-to-day operations through shared responsibility, automation and feedback. Its aim is to improve how changes are delivered and run, not just to release more often.
Turn understanding into practice
Each entry includes an exercise and related concepts. For structured learning with an instructor, explore the course catalogue and choose a programme that fits your starting point.
Explore training courses →