THE DATA & AI
DICTIONARY
85 clear definitions of Data Science, Machine Learning, Data Engineering and Artificial Intelligence concepts.
A
A2A Protocol
The A2A (Agent-to-Agent) protocol is an open standard proposed by Google in April 2025 to let AI agents from different providers communicate and collaborate. Complementary to MCP (Model Context Protocol), it standardizes the discovery, negotiation, and exchange of tasks between agents.
Read the definition →Agentic Analytics
Agentic Analytics refers to systems where autonomous AI agents explore data, break a complex question down into dozens of queries, and produce governed analysis with minimal human oversight rather than manually steering every step.
Read the definition →AI Act
The AI Act is the European regulation governing the development and use of artificial intelligence systems within the EU, with obligations scaled to each system's level of risk.
Read the definition →AI Agents
An AI agent is an artificial intelligence system capable of perceiving its environment, reasoning, and acting autonomously to achieve goals, without constant human intervention.
Read the definition →AI Champion
An AI Champion is an employee trained to accelerate AI adoption within their team or department. They bridge technical expertise and business needs, identify relevant use cases, and help colleagues get up to speed with AI tools.
Read the definition →AI Copilot
An AI copilot is an AI assistant embedded in a business tool that boosts a user's productivity by suggesting, correcting, and automating tasks within their workflow. Unlike an autonomous AI agent, a copilot assists a human who keeps control of decisions.
Read the definition →AI Death Cycle
The AI Death Cycle is the vicious circle in which AI projects fail for lack of solid data foundations, deepening the organization's skepticism, reducing investment, and worsening the odds of success for the next projects. Understanding this cycle is the first step to breaking out of it.
Read the definition →AI Governance
AI Governance refers to the organizational framework, policies, processes, roles, and committees, that governs the development and use of AI within an organization. It covers risk management, regulatory compliance, ethics, and the transparency of AI systems.
Read the definition →AI Product Manager
The AI Product Manager is a profile that combines product management skills with a deep understanding of AI and machine learning. They drive the design of AI-powered products, bridging technical teams, the business, and users.
Read the definition →AI Readiness
AI Readiness measures an organization's ability to adopt and effectively use artificial intelligence. It assesses maturity across several dimensions: data quality, in-house skills, company culture, technical infrastructure, and governance.
Read the definition →Analytics Engineer
The Analytics Engineer applies Software Engineering best practices (tests, CI/CD, versioning) to data transformation, bridging the gap between the Data Engineer and the Data Analyst.
Read the definition →AWS
AWS (Amazon Web Services) is Amazon's cloud computing platform, the global market leader with over 200 services spanning compute, storage, databases, AI, and machine learning.
Read the definition →Error Analysis
Error analysis is a systematic method for identifying, categorizing, and fixing a Machine Learning model's weaknesses. It guides improvement iterations by targeting the subpopulations where the model fails most.
Read the definition →Transformer Architecture
The Transformer architecture is a type of neural network based on the attention mechanism, introduced by Google in 2017. It is the technical foundation of every major large language model today (GPT, Claude, Llama, Mistral) and has transformed natural language processing.
Read the definition →B
Batch Ingestion
Batch ingestion processes data in batches at regular intervals, as opposed to real-time streaming. It's the dominant pattern in modern data pipelines.
Read the definition →Vector Database
A vector database is a storage system optimized to index and search embedding vectors. It enables semantic search by finding the closest items in a mathematical space, and is a foundational building block of RAG architectures.
Read the definition →C
Caveman
Caveman is a utility that reduces the verbosity of Claude Code's responses by pushing it to express itself tersely and simply, caveman-style. Where RTK reduces input tokens, Caveman targets output tokens, without touching the model's internal reasoning.
Read the definition →ChatGPT
ChatGPT is the conversational AI application developed by OpenAI, built on the GPT models. Its launch in late 2022 triggered the mass adoption of generative AI among the general public and in business.
Read the definition →Citizen Data Scientist
A Citizen Data Scientist is a business professional able to use data analysis and machine learning techniques through no-code/low-code tools, without specialized data science training.
Read the definition →Claude Agent SDK
The Claude Agent SDK provides the building blocks for building your own custom version of Claude Code or adding agentic behavior to an AI application: creating agents, overriding prompts, adding new tools. It's built on the same core as Claude Code, with access to skills, dynamic workflows, and more.
Read the definition →Cloud
Cloud Computing gives access to computing resources (servers, storage, databases, AI) over the internet, without owning or managing the underlying physical infrastructure.
Read the definition →Compound AI Systems
A Compound AI System is an AI system made up of several orchestrated components (LLMs, retrievers, tools, code) to solve complex tasks. Rather than relying on a single model for everything, this modular approach combines each component's strengths to get better results.
Read the definition →Context Engineering
Context engineering is the practice of designing and optimizing the entire context provided to an LLM to get reliable answers. Beyond the prompt itself, it covers dynamic information selection, memory management, and the orchestration of data sources.
Read the definition →CRISP-ML
CRISP-ML(Q) is a methodology that standardizes the Machine Learning project lifecycle into 7 steps, from problem understanding to production monitoring, with a quality lens throughout.
Read the definition →D
Data Analyst
The Data Analyst collects, cleans, and analyzes company data to produce the metrics, reports, and recommendations that inform strategic and operational decisions.
Read the definition →Data As A Product
Data as a Product is an approach that treats data as a product in its own right, with an owner, users, and quality standards. It's one of the 4 pillars of Data Mesh.
Read the definition →Data Business Model Canvas
The Data Business Model Canvas is a framing tool that adapts Alex Osterwalder's Business Model Canvas to data projects, structuring the thinking around 9 data-centric sections.
Read the definition →Data Contract
A Data Contract is a formal agreement between a data producer and a data consumer that defines the structure, format, expected quality, and SLAs of the data being exchanged. It formalizes each party's commitments and makes data dependencies explicit and verifiable.
Read the definition →Data Drift
Data Drift refers to the change in the distribution of a Machine Learning model's input data over time, which can degrade its performance even though the model itself hasn't changed.
Read the definition →Data Engineer
The Data Engineer designs, builds, and maintains the data pipelines and infrastructure that let organizations collect, transform, and make their data available at scale.
Read the definition →Data Fresco
The Data & AI Fresco is a collaborative workshop created by Hymaïa that raises participants' awareness of data and artificial intelligence challenges through a playful, interactive format.
Read the definition →Data Governance
Data Governance is the strategic and operational framework that defines the rules, roles, and processes for managing an organization's data reliably, securely, and in compliance with regulations.
Read the definition →Data Lineage
Data Lineage traces the complete journey of data within an organization: its origin, the transformations it undergoes, and the systems it passes through, from source to final consumption.
Read the definition →Data Literacy
Data Literacy refers to the ability to read, understand, analyze, and communicate with data. It's a cross-functional skill needed across every role.
Read the definition →Data Mesh
Data Mesh is a decentralized data architecture paradigm built on 4 pillars: domain-based ownership, Data as a Product, a self-service platform, and federated governance. It's the application of Domain-Driven Design to data.
Read the definition →Data Pitfalls
Data pitfalls are the recurring traps that keep organizations from leveraging their data at scale. They fall into three categories: organizational, methodological, and technical.
Read the definition →Data Platform
A Data Platform is the set of tools, services, and infrastructure that let an organization collect, store, transform, analyze, and distribute its data. It's the technical foundation of any data strategy.
Read the definition →Data Product Manager
The Data Product Manager drives the strategy and roadmap for data-driven products. They combine product expertise, an understanding of data, and business vision to maximize the value created.
Read the definition →Data Steward
The Data Steward is the guardian of data quality within an organization. Responsible for the business glossary and the Data Catalog, they're the first point of contact for every data user.
Read the definition →Data Storytelling
Data Storytelling is the art of turning data into compelling narratives. By combining data, narrative, and visualization, it makes analysis accessible and actionable for any audience.
Read the definition →Data Strategist
The Data Strategist defines and drives a company's data strategy. They align data initiatives with business objectives to maximize the value created from data.
Read the definition →dbt
dbt (data build tool) is an open source data transformation tool that lets Analytics Engineers apply software engineering best practices (versioning, tests, documentation) to their SQL pipelines.
Read the definition →Synthetic Data
Synthetic data is artificially generated data that reproduces the statistical properties of real data without containing personal information. It's used to train AI models, test systems, or share datasets while respecting privacy.
Read the definition →E
AI Evaluation
AI evaluation refers to the methods and metrics used to measure the quality and reliability of AI systems, LLMs in particular. It covers benchmarks, automated evaluation (LLM-as-a-judge), human evaluation, and RAG-specific metrics.
Read the definition →Embedding
An embedding is a numerical representation of a piece of data (text, image, audio) as a vector in a mathematical space. Embeddings capture the semantic meaning of data and measure similarity between items, which makes them essential to RAG and semantic search.
Read the definition →F
Feature Store
A Feature Store is a platform that centralizes the storage, management, and sharing of the features used to train and serve Machine Learning models, ensuring consistency between training and production.
Read the definition →Fine-tuning
Fine-tuning is the process of adapting a pre-trained language model (LLM) to a specific domain or task by retraining it on a targeted dataset. It produces a specialized model without bearing the cost of a full training run.
Read the definition →G
/goal
/goal is a Claude Code command that applies declarative programming to AI agents: you define a measurable objective rather than a sequence of instructions. An evaluator judges each iteration, feeds the latest failure reason back to the agent, and restarts the cycle until the objective is met.
Read the definition →Generative AI
Generative AI refers to artificial intelligence systems capable of creating original content (text, image, code, audio) from natural language instructions. It relies mainly on LLM and Transformer-type models.
Read the definition →GEO (Generative Engine Optimization)
GEO (Generative Engine Optimization) is the optimization of content to be selected and cited by generative search engines such as Google AI Overviews, Perplexity, or ChatGPT Search. It's the evolution of classic SEO for the generative AI era.
Read the definition →GraphRAG
GraphRAG combines knowledge graphs and RAG to let LLMs reason over complex relationships between entities. By structuring data as a graph before retrieval, this approach improves answers to questions requiring several logical hops.
Read the definition →Grounding
Grounding is the technique of anchoring an LLM's responses in verifiable data sources. By connecting the model to documents, databases, or APIs, grounding reduces hallucinations and lets generated claims be sourced.
Read the definition →Guardrails (AI)
AI guardrails are control mechanisms applied to the inputs and outputs of LLMs to guarantee the safety, compliance, and quality of responses. They filter inappropriate content, validate formats, detect manipulation attempts, and check factual consistency.
Read the definition →I
Agentic AI
Agentic AI refers to a category of artificial intelligence systems capable of planning, reasoning, and acting autonomously to reach a goal. Unlike an LLM used conversationally, an agentic system makes decisions, uses tools, and iterates without human intervention at every step.
Read the definition →AI Ops
AI Ops refers to the set of practices, and the dedicated role, focused on structuring, accelerating, and steering the adoption of artificial intelligence across an organization.
Read the definition →Multimodal AI
Multimodal AI refers to AI systems capable of processing and generating several types of data, text, images, audio, video, in an integrated way. Unlike models specialized in a single modality, multimodal models understand and connect information across formats.
Read the definition →Sovereign AI
Sovereign AI refers to a state or economic zone's ability to develop, host, and control its own AI systems without depending on foreign actors. In Europe, this challenge covers models, compute infrastructure, data, and skills.
Read the definition →K
Knowledge Graph
A knowledge graph is a data structure that organizes information as entities connected by semantic relationships. Used to structure an organization's knowledge, it's seeing renewed interest thanks to generative AI and GraphRAG.
Read the definition →Knowledge Infrastructure
Knowledge Infrastructure refers to the systems, processes, and practices that capture, structure, and distribute an organization's knowledge. It includes knowledge graphs, vector databases, taxonomies, and curation processes.
Read the definition →L
/loop
/loop is a Claude Code command that repeats a prompt at a regular interval within the same session: you specify a duration (5 seconds, 5 minutes, 10 minutes...) and an instruction, which is re-run automatically at each interval for up to 7 days.
Read the definition →LLM (Large Language Model)
An LLM (Large Language Model) is an artificial intelligence model trained on vast amounts of text, capable of understanding and generating natural language. GPT, Claude, Llama, and Mistral are examples of LLMs powering generative AI applications.
Read the definition →LLMOps
LLMOps refers to the engineering practices for deploying, monitoring, and maintaining LLM-based applications in production. An extension of MLOps adapted to large language models, it covers prompt management, output evaluation, cost tracking, and governance.
Read the definition →Loop Engineering
Loop engineering is an evolution in working with AI agents: rather than prompting them manually at every step, you let them orchestrate themselves in a loop, with access to skills, tools, and persistent memory. The developer stays at the center, in charge of the skills and of verification.
Read the definition →M
MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open source protocol created by Anthropic that standardizes the connection between LLMs and external data sources or tools. It plays the same role for AI that USB played for peripherals: a universal interface.
Read the definition →ML Engineer
The ML Engineer designs, trains, and deploys machine learning models in production. Halfway between a Data Scientist and a software engineer, they turn prototypes into reliable, scalable systems.
Read the definition →MLOps
MLOps applies DevOps principles to machine learning: deployment automation, monitoring of models in production, and management of the full lifecycle, from training to retraining.
Read the definition →MLOps Engineer
The MLOps Engineer automates the machine learning model lifecycle in production: training pipelines, deployment, monitoring, and data quality. They're the DevOps of ML.
Read the definition →Modern Data Stack
The Modern Data Stack refers to the set of cloud-native tools used to collect, store, transform, and analyze data. It replaces monolithic architectures with specialized, interoperable building blocks.
Read the definition →O
AI-Native Organization
An AI-native organization is a company designed, or transformed, to operate with AI at the core of its processes, products, and culture. Beyond adopting tools here and there, it rethinks its ways of working, its roles, and its governance around AI capabilities.
Read the definition →Multi-Agent Orchestration
Multi-agent orchestration refers to the coordination of several specialized AI agents working together to solve complex tasks. Each agent has its own role and tools, while an orchestrator manages the execution flow, delegation, and synthesis of results.
Read the definition →P
Poetry (Python)
Poetry is a dependency management and packaging tool for Python. It unifies dependency declaration, virtual environment creation, and package publishing into a single workflow.
Read the definition →Product Builder
The Product Builder is a new archetype of product manager who uses AI and no-code tools to prototype, test, and iterate directly, without systematically relying on a development team. It embodies the evolution of the PM role in the age of generative AI and vibe coding.
Read the definition →Product Manager
The Product Manager defines a product's vision, prioritizes features, and coordinates teams to deliver value to users. With the rise of data and AI, the role now incorporates new skills.
Read the definition →Prompt Engineering
Prompt engineering is the discipline of crafting precise, structured instructions to get the best results from an LLM. It brings together a set of techniques, few-shot, chain-of-thought, system prompts, to guide the model's behavior.
Read the definition →R
RAG (Retrieval-Augmented Generation)
RAG (Retrieval-Augmented Generation) is an architecture that enriches an LLM's responses by feeding it relevant documents retrieved from a knowledge base. This approach reduces hallucinations and lets you leverage internal data without retraining the model.
Read the definition →Responsible AI
Responsible AI refers to the set of practices that ensure artificial intelligence systems are developed and deployed ethically, fairly, transparently, and securely.
Read the definition →RTK
RTK (Raw Token Killer) is a utility that reduces the number of tokens sent into an AI agent's context by filtering the raw output of about a hundred common commands like read, ls, cat, npm test, or go test. It can cut token usage on these commands by up to 90%, without necessarily cutting the bill by the same amount.
Read the definition →Semantic Search
Semantic search is an information retrieval method based on the meaning of words rather than exact matching. By using embeddings to represent texts and queries as vectors, it finds relevant documents even when the wording differs. It's the key infrastructure behind RAG.
Read the definition →S
Semantic Layer
The Semantic Layer is a logical layer that translates raw data into shared business concepts, so that every tool consuming the data (BI, application, AI agent) returns exactly the same result.
Read the definition →Shadow AI
Shadow AI refers to the unauthorized or unmanaged use of AI tools by an organization's employees, outside of approved systems. Analogous to Shadow IT, it exposes the company to data leakage risks, regulatory non-compliance, and decisions based on unreliable results.
Read the definition →Spark
Apache Spark is an open source distributed computing framework designed for large-scale data processing. It stands out for its in-memory execution, which makes it far faster than Hadoop MapReduce.
Read the definition →