Getting started with Data Architect

Under Data module, you can find the Data Architect Agent button in the top right corner across all tabs. When you open Data Architect, the following welcome screen is displayed:

Two buttons are presented:

• Browse Data: Opens a view of your existing datasets and Business Views. Use this when you want to edit or extend an existing BV, or when you want to explore what data is already loaded before starting a new build.

• Quick Start: Begins a new Kaiya conversation immediately. Kaiya greets you in the chat panel and asks what data you want to work with or what business questions you are trying to answer.

Browse Data

Clicking Browse Data on the welcome screen opens a dialog with two top-level tabs:

• Data Sources: Displays all connected data sources as cards in a grid layout. A search bar at the top right lets you filter by name.

• Datasets: As per the datasources you selected, displays pre-loaded datasets that already exist in Tellius. This tab is used when you want to build a BV from data that has already been loaded into Tellius rather than connecting directly to a live datasource.

• Tables: The Tables tab lets you browse individual tables within a selected data source.

Select the required data sources, datasets, and tables. Click Continue button to proceed. Kaiya conversation screen will be displayed as follows. A similar split-pane interface will be displayed when you click on Quick Start button.

The Data Architect agent accepts file uploads (PDF, images, Excel/CSV, dbt model) during a conversation to provide additional context. Files can be attached using the paperclip icon in the chat input bar. Supported file types and their uses:

In practice, attaching a comprehensive handoff packet at the start of a conversation can significantly accelerate the process. The agent extracts KBQs, join specifications, table lists, and metric definitions from the document and uses them to pre-populate the BV plan, reducing the number of clarification questions needed.

You can then use the + button in the chat input bar (bottom left) to add additional data sources, datasets, or existing BVs to the current conversation without starting a new session. For example, if you began a conversation with two datasets and later realize you need a third table, you can click +, select it from the Data Sources, Datasets, or Tables tab, and add it to the active session.

Edit Business View: This tab displays a list of all existing Business Views in your Tellius instance. You can select one or more BVs and click Continue to open those BVs in Data Architect for editing.

The agent loads all selected BVs, reads their YAML configurations, and lists each one by name with a brief description of its structure. It then asks what you want to do (modify joins, add columns, fix data quality issues, republish, or something else) so you can jump straight into the specific change you need without navigating the BV configuration manually.

Conversation History

The Conversation History panel is accessible by clicking the panel icon (the small rectangle icon) in the top-left corner of the Data Architect interface, next to the back arrow. This panel displays a chronological list of all past Data Architect sessions.

Each entry in the list shows three pieces of information:

• Conversation name

• A colored status label indicating the session's outcome. Observed statuses include draft (gray), approved (green), and executed (green).

• Date and time of when the session last had activity (for example, "12 Mar, 15:11").

A search bar at the top of the panel lets you filter conversations by name. Clicking any entry reopens that conversation in the split-pane interface, loading the chat history and the associated Plan/Diagram/YAML state so you can continue where you left off.

Anatomy of the data architect agent

Each Data Architect session is assigned a conversation name, displayed at the top of the chat panel just below the “Data Architect” header. This name serves as an identifier for the session (for example, “add_missing_dimensions”).

Once a conversation begins, the Data Architect interface is divided into two panels:

• Left panel (Chat): The conversational interface where you interact with the Data Architect agent. You type messages, answer clarification questions, upload files, and approve or reject plans here. A text input at the bottom includes a + button for adding data sources and a paperclip icon for attaching files (PDFs, images, Excel files, data dictionaries).

• Right panel (Plan / Diagram / YAML): A three-tab display that shows the current state of the BV plan. The three tabs are:

Plan

A structured summary of what the agent is proposing: the analysis goal, tables, joins, columns, and expected outcomes. Plans display an execution status badge (such as "Executed" or "Draft").

Plans follow a consistent document structure. When the agent proposes a modification or a new BV build, the plan displayed in the Plan tab is organized into sections. The following are a few of those:

• Problem Statement (identifies the issue or objective)

• Solution (the proposed approach)

• Changes to YAML (the exact modifications that will be applied, including new sources, new joins, new columns, and role assignments)

• Expected Outcome (what the BV will look like after the changes), and Validation (what checks will be run after publishing to confirm the changes resolved the issue)

This structure ensures you can review the full scope of proposed changes before approving.

Diagram

A visual entity-relationship diagram (ERD) that displays all tables, their columns, data types, and join relationships.

• Each table is labeled as Fact or Dim (dimension).

• Join lines show cardinality (N:1, 1:1, etc.) and include a confidence percentage. The diagram supports two layout modes: "From YAML" (which follows the YAML structure exactly) and "Auto Layout" (which rearranges tables for visual clarity). Zoom controls and a fit-to-screen button are available in the bottom-left corner.

YAML

The raw YAML configuration file that defines the entire Business View. This includes source definitions (dataset names, table roles, data source references, column lists with types), model metadata (join specifications, column mappings, calculated fields), and options (freshLoad flags, sync settings). The YAML is displayed in a syntax-highlighted code editor and is directly editable before publishing.