Lakehouse Anatomy
In data engineering, it is common to use three layers to refine the data. Understanding the role of each one is the map that organizes the entire rest of your work.
- Understand the role of Bronze, Silver and Gold
- Knowing what to put, and what not to put, in each layer
- Recognize good layer organization when you look
Think about a mine. The ore comes out of the earth dirty and mixed (Bronze), goes through a first wash that separates what works (Silver), and only then it becomes the polished piece that goes into the showcase (Gold). Your data they follow exactly this path, and each layer has a clear function. This is the medallion architecture, the standard that Nekt recommends call on Bronze, Silver and Gold to organize a lakehouse.
- Faithful mirror of the origin
- No editing, immutable
- Raw type, often text
- One set per source
- Renamed and clear columns
- Correct types, no duplicates
- One table per entity
- Pleasant to read
- Combine sources, apply business rules
- Tables ready for analysis
- This is where BI, MCP and API read from
- A source of truth
Bronze, Silver and Gold is a Nekt recommendation, not an obligation. Each company can organize the layers in its own way whatever makes the most sense to her. What really matters is the concept: raw data, refined data and ready data for analysis. Bronze, Silver and Gold is just the suggested name for these three stages.
Bronze: the given how it arrived
The layer that Nekt suggests calling
Bronze is the exact portrait of
source. Nekt brings data from HubSpot, the bank, Meta Ads, and stores
as it came, without touching. Here the types usually come raw according to the
that each data source uses (a monetary value arrives as text
"1234.50", a date as a string), there are structures
nested, and the column names are what the source uses, not always
friendly. The golden rule: you don't edit Bronze.
It changes only when the source changes.
Silver: clean and reliable data
In the layer Silver the data remains tidy. You type the fields (text becomes a real number and date), renames columns to clear names, removes duplicates, and leaves one table by business entity (customers, orders, subscriptions). It's the first place where the data is really pleasant to consult, but still without combining sources or applying heavy business rules.
Gold: ready-to-consume data
The layer Gold That's where the die turns decision. Here you combine sources (orders + customers + channels), apply the agreed business rule and sets up large tables, ready to analysis. It is from this layer that dashboards, AI agents (via MCP) and the Data API read. A number, a table, a source of truth for everyone.
Nekt's MCP creates queries and notebooks for you to from natural language. Order "turn order Bronze in a clean and typed Silver" or "set up a Gold billing per channel" and it generates transformation. This makes the job MUCH easier structure Bronze, Silver and Gold: you describe what you want, review, and adjusts, rather than writing each step from scratch.
The same line, crossing the three layers
See a single order being refined. Notice how it comes out of something raw and difficult to use for something that any person (or agent) understands in time.
| order | value | status | date | customer | channel |
|---|---|---|---|---|---|
| a83f | R$ 1,234.50 | approved | 08/03/2026 | Silva Store |
The Silver layer is where cleaning happens once and becomes reused by all Gold tables. Skipping this step spreads the Same cleanliness (and same mistakes) everywhere.
Pocket Summary
| Layer | What lives | Format | Who reads | Change when |
|---|---|---|---|---|
| Bronze | Source raw data | How does it come from the source | No one direct | The data source changes |
| Silver | Clean entities | Typed, 1 per entity | Gold Transformations | You improve cleaning |
| Gold | Business Tables | Wide, ready for analysis | BI, MCP, Data API | The business rule changes |
An agency with 12 clients had Meta Ads, Google Ads and CRM each in Bronze, 36 raw sets. In Silver, it standardized each source in a clean table by type. In Gold, he created a table with one line per customer, channel and day. This single Gold table feeds the customer report, the internal dashboard and an agent who answers "which campaign earned the most this week". Three consumptions, a source of truth.