Getting Started with TellR: AI-Powered Slides from Databricks

Generate presentation-ready slides from the Bakehouse sample dataset using TellR

TellR
Databricks Apps
AI
Tutorials
Learn how to install and use TellR, an agentic Databricks App that turns your data into interactive slide decks through natural language.
Modified

31/01/2026

NoteScreenshots & Video

All the code in this tutorial post is rendered against a Databricks workspace before publication. A video walkthrough is also provided for your convenience.

Summary

  • TellR is an open-source Databricks App that generates interactive slide decks from your data using natural language
  • You’ll install TellR, connect it to the Bakehouse sample dataset in Unity Catalog, and create your first presentation
  • We’ll create a custom DailyDatabricks slide style and a profile that ties together a Genie space, style, and deck prompt
  • TellR queries your data through Genie spaces, respecting Unity Catalog permissions automatically
  • Slides are rendered as interactive HTML with Chart.js visualizations and can be exported to PDF

Video Walkthrough

What is TellR?

TellR is an agentic application that generates presentation-ready slides from enterprise data through conversational interaction. It runs as a Databricks App and integrates with Genie spaces to query live, governed data.

Key features:

  • Conversational slide creation — Describe what you want in plain English and TellR builds the deck
  • Genie integration — Queries your data through Genie spaces, so your slides reflect real, up-to-date numbers
  • Unity Catalog security — All data access respects your existing permissions; no data leaks outside governance boundaries
  • Interactive visualizations — Generates Chart.js-based charts that are interactive in the browser
  • Custom slide styles — Define your own branding with colours, typography, and layout rules
  • Profiles — Bundle a Genie space, slide style, and deck prompt into a reusable configuration
  • Iterative editing — Refine slides with follow-up prompts like “add a Q3 comparison” or “change the chart to a bar graph”

Prerequisites

Before you begin, make sure you have:

  • A Databricks workspace with Apps enabled
  • Permission to create a Lakebase (or an existing schema you can use)
  • Access to the Bakehouse sample dataset in Unity Catalog (available by default in most workspaces)
  • On Behalf Of (OBO) authentication enabled for Databricks Apps (see Step 1 for details)

The Bakehouse Dataset

The Bakehouse dataset is a sample dataset available in Unity Catalog. It models a bakery franchise business with sales, customers, suppliers, and reviews.

Table Description
sales_transactions Fact table with individual sales records
sales_customers Customer dimension table
sales_franchises Franchise/store location details
sales_suppliers Product supplier information
media_customer_reviews Customer feedback and ratings
media_gold_reviews_chunked Processed review data for analytics
TipFinding the Dataset

In your Databricks workspace, navigate to the Catalog Explorer and look under samples > bakehouse. If you don’t see it, your workspace administrator may need to enable sample datasets.

Let’s explore the tables to get familiar with the data:

# Browse the bakehouse schema
spark.sql("SHOW TABLES IN samples.bakehouse").show()
+---------+--------------------+-----------+
| database|           tableName|isTemporary|
+---------+--------------------+-----------+
|bakehouse|media_customer_re...|      false|
|bakehouse|media_gold_review...|      false|
|bakehouse|     sales_customers|      false|
|bakehouse|    sales_franchises|      false|
|bakehouse|     sales_suppliers|      false|
|bakehouse|  sales_transactions|      false|
+---------+--------------------+-----------+

# Preview the sales transactions
spark.table("samples.bakehouse.sales_transactions").limit(10).show()
+-------------+----------+-----------+--------------------+--------------------+--------+---------+----------+-------------+----------------+
|transactionID|customerID|franchiseID|            dateTime|             product|quantity|unitPrice|totalPrice|paymentMethod|      cardNumber|
+-------------+----------+-----------+--------------------+--------------------+--------+---------+----------+-------------+----------------+
|      1002961|   2000253|    3000047|2024-05-14 12:17:...|  Golden Gate Ginger|       8|        3|        24|         amex| 378154478982993|
|      1003007|   2000226|    3000047|2024-05-10 23:10:...|Austin Almond Bis...|      36|        3|       108|   mastercard|2244626981238094|
|      1003017|   2000108|    3000047|2024-05-16 16:34:...|Austin Almond Bis...|      40|        3|       120|   mastercard|2490570234487424|
|      1003068|   2000173|    3000047|2024-05-02 04:31:...|         Pearly Pies|      28|        3|        84|         amex| 343808569426192|
|      1003103|   2000075|    3000047|2024-05-04 23:44:...|         Pearly Pies|      28|        3|        84|         visa|4377080942201798|
|      1003147|   2000295|    3000047|2024-05-15 16:17:...|Austin Almond Bis...|      32|        3|        96|         amex| 371093774812677|
|      1003196|   2000237|    3000047|2024-05-07 11:13:...|       Tokyo Tidbits|      40|        3|       120|   mastercard|5538807345848392|
|      1003329|   2000272|    3000047|2024-05-06 03:32:...|     Outback Oatmeal|      28|        3|        84|         visa|4872480716880043|
|      1001264|   2000209|    3000047|2024-05-16 17:32:...|         Pearly Pies|      28|        3|        84|   mastercard|5287105980593305|
|      1001287|   2000120|    3000047|2024-05-15 08:41:...|Austin Almond Bis...|      40|        3|       120|         amex| 376211012259783|
+-------------+----------+-----------+--------------------+--------------------+--------+---------+----------+-------------+----------------+

# Check the schema
spark.table("samples.bakehouse.sales_transactions").printSchema()
root
 |-- transactionID: long (nullable = true)
 |-- customerID: long (nullable = true)
 |-- franchiseID: long (nullable = true)
 |-- dateTime: timestamp (nullable = true)
 |-- product: string (nullable = true)
 |-- quantity: long (nullable = true)
 |-- unitPrice: long (nullable = true)
 |-- totalPrice: long (nullable = true)
 |-- paymentMethod: string (nullable = true)
 |-- cardNumber: long (nullable = true)

Step 1: Install TellR

Open a new notebook in your Databricks workspace and run the following cells:

%pip install --upgrade databricks-tellr databricks-sdk==0.73.0
dbutils.library.restartPython()

Then create the TellR app:

import databricks_tellr as tellr

tellr.create(
    lakebase_name="tellr-db",
    schema_name="app_data",
    app_name="tellr",
    app_file_workspace_path="/Workspace/Users/you@example.com/.apps/tellr"
)
NoteWorkspace Path

Replace you@example.com with your actual Databricks username. This is the workspace path where TellR’s application files will be stored.

You may encounter this error during deployment:

Deployment failed: Databricks Apps - user token passthrough feature
is not enabled for organization <org_id>.
Config: host=<workspace_url>, auth_type=runtime

This means On Behalf Of (OBO) authentication is not enabled for your workspace. Databricks Apps require this feature so the app can act on behalf of the logged-in user when accessing data through Unity Catalog.

To resolve this, a workspace administrator needs to enable the On Behalf Of auth setting for Databricks Apps. See the screenshot below for where to find this in the admin console.

Enabling On Behalf Of authentication for Databricks Apps

Installing TellR in a Databricks notebook

Step 2: Verify the Lakebase

After installation completes, TellR creates a Lakebase (a managed PostgreSQL database) to store its application data. You can verify it was created successfully by navigating to SQL Warehouses in your workspace and checking for the tellr-db instance.

The tellr-db Lakebase is created and available

The Lakebase should show a status of Available with PostgreSQL version 16. This is where TellR stores profiles, slide styles, history, and other app configuration.

Step 3: Authorize the App

When you first open TellR, you’ll be prompted to authorize the app to act on your behalf. This is the On Behalf Of (OBO) authentication flow — TellR needs permission to read Unity Catalog tables and manage model serving endpoints under your identity.

Permission request when first opening TellR

Review the requested permissions:

  • Read Unity Catalog tables — so TellR can query your data through Genie spaces
  • Read Unity Catalog schemas and catalogs — for data discovery
  • Manage your model serving endpoints — TellR uses a Foundation Model endpoint for slide generation

Click Authorize to proceed.

Step 4: Explore TellR

Once authorized, you’ll land on the TellR help page. This gives you an overview of the app’s features and navigation.

TellR landing page

The top navigation bar provides access to:

  • Generator — The main chat interface for creating slides
  • History — Previously generated slide decks
  • Profiles — Configuration profiles that bundle data sources, styles, and prompts
  • Deck Prompts — Pre-built prompt templates for specific presentation types
  • Slide Styles — Visual styling rules for slide generation
  • Help — The overview page you’re looking at

Before we can generate slides, we need to set up three things: a Genie space for our data, a custom slide style, and a profile that ties them together.

Step 5: Create a Genie Space

TellR queries your data through Genie spaces. We need to create one for the Bakehouse dataset.

  1. In your Databricks workspace, go to Genie in the left sidebar
  2. Click New Genie space
  3. In the Connect your data dialog, navigate to All catalogs > samples > bakehouse

Navigating to the Bakehouse schema in the Genie space creator
  1. Select the tables you want TellR to query. For this tutorial, select sales_transactions, sales_suppliers, and sales_customers:

Selecting Bakehouse tables for the Genie space
  1. Click Create to save the Genie space
TipGenie Space Tips

The more descriptive your Genie space instructions are, the better TellR will understand your data. You can add a description like: “Sales and customer data for a bakery franchise business. Includes transactions, customer details, and supplier information.”

Step 6: Create a Custom Slide Style

Before creating a profile, let’s set up a custom slide style. TellR comes with a System Default style, but you can create your own to match your branding.

  1. In TellR, click Slide Styles in the top navigation
  2. Click + Create Style
  3. Give it a name — we’ll use DailyDatabricks for this example
  4. Define the style content with your brand colours, typography, spacing, and chart palette

Creating a custom DailyDatabricks slide style

The style content is a text-based specification that controls typography, colours, layout, spacing, and data visualisation palettes. TellR’s AI uses this to generate slides that match your brand identity.

Step 7: Create a Profile

A profile bundles together a Genie space, slide style, and deck prompt into a reusable configuration. The profile creation wizard walks you through five steps.

Step 7.1: Basic Info

Click Profiles in the top navigation, then + Create Profile. Give your profile a name — we’ll call ours BakeHouse.

Step 1: Enter profile name and description

Step 7.2: Select Genie Space

Choose the Genie space you created in Step 5. This tells TellR where to pull data from when generating slides.

Step 2: Select the Genie space for your profile
TipData Description

Fill in the Data Description (for AI Agent) field to give TellR’s AI context about what your data represents. This helps it generate more relevant slide narratives.

Step 7.3: Select Slide Style

Choose the DailyDatabricks style we created in Step 6. This controls the visual appearance of generated slides.

Step 3: Select the DailyDatabricks slide style

Step 7.4: Select Deck Prompt

Choose a deck prompt template that matches the type of presentation you want to create. For this example, we’ll select Quarterly Business Review which structures the deck around performance metrics, achievements, challenges, and strategic recommendations.

Step 4: Select a deck prompt template

Step 7.5: Review and Save

Review your profile settings and save. You can always come back to edit these later.

Step 8: Review Your Profile

After creation, you can view your profile details at any time by clicking on it in the Profiles tab. This shows the complete configuration at a glance.

Viewing the BakeHouse profile configuration

Notice the profile shows:

  • Genie Space — linked to the Bakehouse data
  • Slide Style — DailyDatabricks
  • Deck Prompt — Quarterly Business Review
  • AI Infrastructure — the Foundation Model endpoint being used (in this case, databricks-claude-sonnet-4-5 with temperature 0.7)
TipEditing a Profile

Click the Edit button at the top of the profile view to modify any setting. You can change the Genie space, swap styles, or switch deck prompts without creating a new profile.

Step 9: Generate Your Slides

Now for the main event. Make sure your BakeHouse profile is selected in the top-right dropdown, then navigate to the Generator tab.

Type a prompt describing the presentation you want. For example:

Generate the slides to help me understand the current state of the art and our position for the next FY. We need to think prescriptive.

TellR will begin querying your Genie space and generating slides. You can watch the process in the chat panel — it makes multiple calls to query_genie_space to gather the data it needs.

TellR generating slides from Bakehouse data via Genie queries

After a few moments, the slides appear in the preview panel on the right. You can select individual slides, review the generated content, and continue the conversation to refine them.

The generated FY2025 Strategic Position Review presentation

The generated deck includes:

  • A title slide with your branding
  • Executive summary with key metrics
  • Data visualisations pulled directly from Bakehouse
  • Strategic analysis and recommendations
TipIterative Refinement

You don’t need to get your prompt perfect the first time. TellR supports conversational editing — ask it to change colours, swap chart types, add comparisons, or restructure the narrative. Think of it like chatting with a presentation designer.

TipVerify Your Genie Space Queries

You can check the actual queries to your Genie space directly by clicking the database link on a slide that will take you to the Databricks UI. This helps validate and verify the AI generated queries being used in TellR.

The underlying Genie space querying Bakehouse data

Step 10: Export and Share

TellR provides two export options via the Export button in the top-right of the slide preview:

  • HTML — An interactive, self-contained HTML file with live Chart.js visualizations
  • PDF — A static document for traditional distribution
TipExample Output

Here’s the actual HTML slide deck generated from the Bakehouse dataset in this tutorial. You can scroll through and interact with the charts:

You can also download the PDF version of this presentation.

Tips & Tricks

TipMultiple Profiles

Create different profiles for different use cases — one for executive reviews, another for team updates, another for customer-facing decks. Each can have its own style, and deck prompt.

NoteUnity Catalog Permissions

TellR respects your Unity Catalog permissions. If you can’t see certain data in the Catalog Explorer, TellR won’t be able to query it either. This means your presentations are always within your governance boundaries.

Cleanup & Cost Management

ImportantStop Resources to Avoid Costs

Once you’re finished with this tutorial, make sure you stop the following resources to avoid incurring unnecessary costs:

  1. Stop the Lakebase — Navigate to SQL Warehouses in your workspace and stop the Lakebase database (tellr-db) that was created during installation
  2. Stop the Databricks App — Go to Apps, find the tellr app, and stop it
  3. Stop the Genie data warehouse — The Genie space uses an attached SQL warehouse; make sure it is stopped as well

You should also consider deleting the Genie space you created if you no longer need it, as an active Genie space with an attached warehouse can continue to incur compute costs.

References & Further Reading

Back to top