natural-language

Databricks Genie is Databricks’ natural-language analytics interface. It allows business users to ask questions about governed enterprise data in plain English, instead of writing SQL.

Think of Genie as: Business User → Natural Language → Genie → SQL → Databricks Data → Answer

Genie can interpret the question, generate the appropriate SQL, execute it against Databricks data, and return the result—often with a visualization.

Where Genie fits in Databricks

Users

Users
├── Analysts
├── Engineers
└── Scientists

These people don’t necessarily write SQL.

They ask: “How many salespeople do I have per region?”
or: “Who are our best sales reps by number of closed deals?”
or: “What’s the open pipeline for the current quarter?”

So this is the Natural Language interface.

Databricks Genie (Orange box)

Genie is NOT a Database, It is the conversational layer between the user and the Databricks data platform.

GenAI Query Engine

GenAI Query Engine
-------------------------
Intent Analysis
SQL Generation
Debug Insights
Metadata Creation

This is essentially the AI reasoning/query-generation layer behind the Genie experience shown in the diagram.

For example, user asks:

“Who are our best sales reps by number of closed deals?”

The system has to figure out:

Intent:
    Find top sales reps

Metric:
    number of closed deals

Dimension:
    sales rep

Data:
    sales opportunity / sales tables

Query:
    SQL

So:

Natural Language
       │
       ▼
Intent Analysis
       │
       ▼
SQL Generation
       │
       ▼
SQL Query

That’s why I previously said Genie is not simply “ChatGPT that writes SQL.”

There is an AI/query layer that has to understand the business question and map it to available data.

Unity Catalog

Now look carefully at the blue box:

Unity Catalog
-------------------------
Table Metadata
Column Definitions
Data Lineage
Access Controls

How does Genie know what revenue means? It doesn’t magically know your company’s definition. It gets context from the governed data environment. For example:

Table:
sales.orders

Columns:
customer_id
order_date
amount
sales_rep
region

And metadata/business context can tell the system things such as:

amount = sales amount
customer_id → customer
sales_rep → salesperson
region → sales region

It is showing what a Genie Space / Genie Agent actually looks like to the business user.

The title is: Sales Opportunities.

That is the name of this particular Genie Agent/Space.

① “Sales Opportunities”

This is the domain/purpose of the Genie Agent. The description says it provides information about:

sales opportunities
sales reps
regions
sales targets
specific sales opportunities
future sales numbers

② Star

The star next to: Sales Opportunities

is basically marking this Genie Agent as a favorite. Not an architectural component.

③ New chat

This is where the user starts a new conversation. For example:

New chat   ↓"What were our sales in Ontario last quarter?"

Each conversation maintains its own conversational context. Databricks documents that follow-up questions in the same thread can use the previous questions and answers as context.

④ History

This is the conversation history. For example:

History ├── Sales by region ├── Top sales reps ├── Open pipeline └── Q2 revenue

This is not database history. It’s chat/conversation history.

⑤ Data

When you click: Data

you can see the datasets configured for this Genie Agent. For example:

Data ├── sales_opportunities ├── sales_reps ├── customers └── sales_targets

And you can inspect columns and descriptions.

Databricks explicitly documents the Data tab as showing the tables selected for the Genie Agent, including their columns and descriptions.

This is part of the answer to “How does Genie know which table?”

It isn’t looking blindly at every database in the world.

The Agent has a curated set of data assets.

⑥ Monitoring

This is for the people managing the Genie Agent. For example:

User asked:"What is our revenue?"Genie generated:SELECT ...Result:Wrong

The manager can monitor interactions and use feedback to improve the Agent.

Databricks provides monitoring for Genie interactions and quality review.

This is especially important in an enterprise architecture.

You don’t just: Build Genie → Done

You need:

Build
 ↓
Test
 ↓
Monitor
 ↓
Review incorrect answers
 ↓
Improve instructions / SQL examples / semantics
 ↓
Retest

⑦ Share

This allows the Genie Agent/conversation to be shared with other users.

Again, this is a user collaboration feature, not a data-processing component.

The bottom of Image 2 is the most important

You have:

How many salespeople do I have per region?

Then:

Who are our best sales reps by # of closed deal?

Then:

Which mid market accounts have generated the most sales globally?

Then:

What's the open pipeline for the current quarter?

These are natural-language business questions.

The user doesn’t know or care whether the underlying SQL is:

SELECT ...FROM ...JOIN ...GROUPBY ...

The user simply asks the business question.

Databricks Genie

Open Azure Databricks → see Genie → click it → open “Bakehouse Sales Starter Space”(for example)

The more accurate sequence is:

1. Open your Azure Databricks workspace

→ 2. Find/access Genie Agents

→ 3. You will see the Genie Agents/Spaces that you have access to

→ 4. If a demo/sample Genie Agent such as the Bakehouse example has been provided in that environment, you can open it

→ 5. Click it and you get the Genie chat interface

→ 6. Ask a natural-language question, for example:

“What were the top-selling products last month?”

Genie then uses the configured data sources, instructions, example SQL, etc., to generate an analytical query and return the result.

Also, Genie itself is not your data. A Genie Agent is configured against specific Unity Catalog tables/views and requires a SQL Warehouse to execute the queries.

The key distinction

Bakehouse Sales Starter Space = an example Genie Agent

Genie = the capability/product

Your company’s Genie Agent = a Genie Agent you configure for your own business data

So I should not have described “Bakehouse Sales Starter Space” as though it were the normal entry point to Genie in every Databricks workspace.

Example

Go to Azure Databricks

Genie Agent data

Genie Agent data MUST BE REGISTERED to Unity Catalog.

Genie
   │
   ├── Azure SQL Database ❌
   ├── On-prem SQL Server ❌
   ├── Oracle ❌
   └── Databricks tables ✅

External DB
     │
     │ ingestion / federation / integration
     ▼
Databricks / Unity Catalog
     │
     ▼
Genie

note: Data Federation. we discussed data federation in Azure Databricks Unity Catalog section. you can read the article from here .

Genie vs RAG

Genie is not the same as RAG. This distinction is important for your AI/LLM architecture learning.

Technology	Main purpose
Genie	Ask questions about structured enterprise data
RAG	Ask questions about unstructured knowledge/documents
LLM Agent	Reason + use tools/APIs to accomplish tasks
Genie + AI/LLM	Natural-language interface for data analytics

e.g.

Genie:

User asks: “Show me revenue by customer for Q2.”
→ generates SQL → queries tables → returns results.

RAG：

user asks: “What is our company’s refund policy?”

→ retrieves relevant documents → sends context to LLM → generates answer.

What makes Genie different from simply using ChatGPT + SQL?

The important enterprise piece is data context and governance.

Genie can be configured around a specific Genie Space, where you provide the relevant data assets and business context.

For example:

Genie Space: Sales Analytics
       │
       ├── sales.orders
       ├── sales.customers
       ├── sales.products
       ├── Business terminology
       ├── Sample questions
       └── Instructions
                │
                ▼
             Genie
                │
                ▼
        Natural Language → SQL

This helps Genie understand things such as:

What does “revenue” mean?
Which table contains sales?
How are customers joined?
What does “active customer” mean?
Which date column should be used?
What metrics are authoritative?

Why this matters for your Databricks Engineer / Architect profile

For you, Genie should be understood as an AI consumption layer on top of the Databricks Lakehouse, rather than another data-engineering pipeline.

Your architecture can be viewed as:

                 ┌──────────────────────┐
                 │ Business Users       │
                 └──────────┬───────────┘
                            │
                    Natural Language
                            │
                            ▼
                 ┌──────────────────────┐
                 │ Databricks Genie     │
                 │ NL → SQL Analytics   │
                 └──────────┬───────────┘
                            │
                            ▼
                 ┌──────────────────────┐
                 │ Databricks SQL       │
                 │ Warehouse            │
                 └──────────┬───────────┘
                            │
                            ▼
                 ┌──────────────────────┐
                 │ Unity Catalog        │
                 │ Governance / Access  │
                 └──────────┬───────────┘
                            │
                            ▼
                 ┌──────────────────────┐
                 │ Lakehouse            │
                 │ Delta Tables         │
                 └──────────────────────┘