Buzzwords, decoded. The words people use as if you already know them, explained simply.
| Term | Simple explanation |
|---|---|
| ETL | Extract, Transform, Load: pull data from a source, clean and reshape it, then load it into its destination, in that order |
| ELT | Extract, Load, Transform: load raw data in first, then transform it inside the warehouse. The modern default now warehouses are cheap and powerful enough to do the heavy lifting |
| Data pipeline | The automated path data takes from a source system to somewhere useful, usually a warehouse or dashboard |
| Data lake | A storage system that holds raw data in its original format, structured or not, until you need it |
| Data warehouse | A storage system built specifically for fast queries over structured, already-cleaned data |
| Data lakehouse | A hybrid: lake-style cheap storage with warehouse-style query performance and structure, made possible by table formats like Iceberg |
| Medallion architecture | Organising data into bronze (raw), silver (cleaned) and gold (business-ready) layers as it moves through a pipeline |
| CDC (Change Data Capture) | Capturing only the rows that changed in a source database, instead of reprocessing the whole table every time |
| Data mesh | An organisational model where each team owns and publishes its own data as a product, instead of one central team owning everything |
| Data contract | A formal agreement on what a dataset looks like and guarantees about it (schema, freshness, quality), so downstream consumers don't get broken silently |
| Reverse ETL | Sending data from the warehouse back out into operational tools (like a CRM), so it can drive actions, not just reports |
| Term | Simple explanation |
|---|---|
| Schema | The structure of a database: what tables exist, what columns they have, and how they relate |
| Star schema | A modelling pattern with one central "fact" table (events, transactions) surrounded by "dimension" tables (who, what, when) that describe it |
| Normalisation | Organising data to reduce duplication, usually by splitting it into more, smaller related tables |
| Idempotency | Running the same operation multiple times gives you the same result as running it once. Essential for pipelines that might retry after a failure |
| Partitioning | Splitting a large table into smaller physical chunks (often by date) so queries only need to scan the relevant slice |
| Term | Simple explanation |
|---|---|
| LLM (Large Language Model) | The type of AI model behind tools like ChatGPT and Claude, trained on huge amounts of text to predict and generate language |
| RAG (Retrieval-Augmented Generation) | Giving an AI model relevant documents to reference at the moment it answers, instead of relying only on what it learned during training |
| Vector database | A database built to store and search "embeddings" (numerical representations of meaning), used to power AI search and RAG |
| MCP (Model Context Protocol) | An open standard that lets AI agents connect to and pull data from tools and databases at runtime, instead of that data being hardcoded |
| Agentic AI | AI systems that don't just answer one question, but take a goal, plan the steps, and carry them out, checking their own progress along the way |
| Semantic layer | A single place where business metrics and definitions are defined once, so every dashboard and tool that uses them agrees |
| Term | Simple explanation |
|---|---|
| CI/CD | Continuous Integration/Continuous Deployment: automatically testing and shipping code changes rather than doing it manually |
| Orchestration | Scheduling and managing the order pipeline steps run in, including retries when something fails |
| Observability | Being able to tell what's actually happening inside a system, pipeline or application, from the outside, through logs, metrics and alerts |
| Data quality | Whether data is accurate, complete and trustworthy enough to make decisions on |
| Data catalogue | A searchable inventory of what data exists across a company, who owns it, and where it came from |