Maia Glossary

An AI pipeline builder uses AI, not drag-and-drop configuration or manual code, to build and maintain data pipelines. See the three kinds of "pipeline builder" on the market, what separates them, and how Maia works as one.

An automated data pipeline runs end-to-end, trigger, move, transform, load, monitor, and increasingly recover, without manual intervention. See the components that make a pipeline automated and the maturity levels between basic scheduling and self-healing.

The use of Agentic AI to plan, build, and optimize data pipelines. Learn how it moves beyond simple automation to provide intelligent, goal-oriented execution, and how Maia uses safe abstractions to reduce manual toil.

Bounded Contexts are the boundary lines that stop a data model from collapsing in on itself. They keep "customer" meaning one thing in Billing and another in Marketing, without forcing the business to pick. Get them right and your architecture scales. Get them wrong and every change becomes a six-month risk assessment.

Context Engineering stops AI from hallucinating. It's the infrastructure that designs what facts an LLM holds when it answers, not just how you ask the question.

An AI agent scoped specifically to data engineering work: building pipelines, catching drift, and fixing failures under human approval. See how it differs from a general AI agent and a multi-agent system.

Data connectors are the pre-built integration components that move data between a source system (a SaaS API, database, or file store) and a target destination such as a cloud data warehouse.

DDL is the part of SQL that defines the structure of a database rather than the data inside it. CREATE, ALTER, and DROP are DDL commands. Get the schema wrong at this layer and every downstream pipeline inherits the mistake.

The automated coordination of complex data workflows, managing dependencies and scheduling across systems so data flows reliably from source to destination.

The planning decisions that happen before a pipeline is built: load pattern, transformation approach, and failure handling. Explore common design patterns and where skipped decisions turn into technical debt.

Data pipeline monitoring tracks whether pipeline runs succeed, how long they take, and where failures happen. See what it covers, how it differs from data observability, and where teams get it wrong.

The coordination layer that decides when pipelines run, in what order, and what happens when one fails. Covers scheduling, dependencies, retries, and how orchestration works in an agentic environment.

Data warehouse migration moves data, schemas, and pipeline logic from one warehouse platform to another. See what actually moves, why these projects run long, and what changes when agents take on the rewrite.

An architectural standard where data storage and computational processing are separated into independent modules. This design allows organizations to scale storage capacity and processing power separately, ensuring they only pay for the specific resources needed for any given workload.

Deterministic AI is logic-driven automation that follows fixed, predefined rules, producing the same outcome every time for the same input. Unlike probabilistic Generative AI, deterministic systems operate with complete transparency, predictability, and auditability, making them ideal for critical workflows such as data pipelines, lineage generation, financial checks, and infrastructure automation.

ELT is the cloud-native data integration standard, extract your data, load it raw into your warehouse, then transform it there using the warehouse's own compute power. Learn how it differs from traditional ETL, why it's faster, and how Maia's AI agents are now automating the entire process.

ETL automation is what lets a pipeline run on its own, on schedule, instead of someone kicking it off by hand every time. Scheduling and connectors have been automated for over a decade. What's new is agents automating the judgment calls too, like deciding what to do when a source schema changes overnight.

ETL process optimization is the work of making extract, transform, load jobs faster and cheaper without rewriting them from scratch. Most slow ETL jobs aren't slow because the tooling is bad. They're slow because nobody fixed the load pattern once the data grew.

A collaborative workflow where AI systems perform the heavy lifting but operate within a transparent environment that allows for human verification. It bridges the gap between the speed of automation and the reliability of expert oversight.

A data design pattern that organizes data into three layers (Bronze, Silver, and Gold) to incrementally improve quality and structure as it flows through a lakehouse. It provides a blueprint for building reliable pipelines on ACID-compliant lakehouse storage.

An AI architecture where specialized agents (like Engineers, Analysts, and Ops) collaborate to solve complex data problems autonomously, rather than relying on a single, rigid script.

An AI technique that lets models solve complex problems by interleaving reasoning (thinking) with acting (tool use).

Reverse ETL is the process of syncing enriched data from a central data warehouse back into operational tools like CRMs, marketing platforms, and support systems. It enables data activation by turning static warehouse insights into timely business actions within the applications teams use every day.

Schema drift occurs when the structure of a data source changes unexpectedly, such as added columns, removed fields, or modified data types, causing traditional pipelines to break or silently ingest incorrect data.

A semantic layer is an abstraction layer that sits between complex data sources and business users, translating technical database schemas into clear, consistent business terminology. It allows non-technical teams to query data using terms like "Annual Recurring Revenue" instead of navigating raw SQL tables and joins.

Serverless data integration is a cloud-based architectural pattern where data pipelines are built, executed, and scaled without requiring manual management of servers or infrastructure. Pipelines scale automatically while engineers focus on delivering outcomes, not keeping the lights on.

A set of strategies for handling changes to dimension attributes over time in a data warehouse. SCD techniques (most commonly Types 1, 2, and 3) determine whether you overwrite history, preserve it, or partially track it.

A dimensional modeling pattern that extends a star schema by normalizing its dimension tables into multiple related tables. The result resembles a snowflake: a central fact table with dimensions branching outward into hierarchical sub-tables.