What the data architect agent can do?

The Data Architect agent orchestrates multiple internal components, each responsible for a specific function in the BV creation workflow:

Function	Key Behaviors
Table profiling	Column type detection (dimension/measure/date/ID), primary and foreign key inference by naming pattern and cardinality, data quality profiling (nulls, uniques, distributions), snapshot vs event table classification
Relationship detection	Name-similarity and type matching across tables, cardinality analysis (1:1, 1:N, N:1, M:N), many-to-many detection with fix suggestions, composite key identification, join intent selection (left outer vs inner)
Formula creation	Key business question to formula translation, row-level and report-level calculation support, natural language refinement loop, Kaiya-only vs general calculated column differentiation
External semantic enrichment	Imports from dbt model files, CSV data dictionaries, SQL comments, PDFs, and images. Treats imported metadata as source-of-truth; LLM only fills gaps
Pre-publish checks	Null analysis, data type compatibility, join key compatibility, cardinality checks, primary key uniqueness, row count integrity, join explosion detection, double counting detection, fan trap and chasm trap detection

Multi-plan and Multi-BV support from a single conversation

For ambiguous requests where multiple valid approaches exist, the agent can generate multiple alternative plans and present them side by side. Each plan represents a different join strategy or aggregation level.

For example, a request for "sales performance by rep and territory" might produce two plans: Plan A joins at transaction grain with a rep-territory hierarchy, while Plan B pre-aggregates at rep-month grain. The agent describes the trade-offs of each approach, and you select which plan to proceed with. Plans are editable in Draft state before execution.

A single Data Architect conversation can produce multiple Business Views representing different analysis perspectives from the same set of source tables. For instance, you might create both a summary BV (aggregated at the monthly level for dashboard use) and a transaction-grain detail BV (for root cause analysis and drill-downs) in the same session.

Editing an existing Business View

Data Architect is not limited to creating new Business Views. You can also use it to modify an existing BV. When you open an existing BV in Data Architect, the agent loads the current YAML configuration and presents the current structure in the right panel.

The agent greets you with a message and then summarizes the current BV, including the table name, the metrics it tracks, and the dimensions it covers.

What you can do in edit mode

• Add new dimension or fact tables to the BV

• Add, modify, or remove joins between tables

• Create new calculated columns or modify existing formulas

• Import metadata from external files to enrich column descriptions

• Run validation checks to identify and fix data quality issues

• Resolve problems surfaced by validation (duplicate rows, formula errors, missing joins)

• Republish the updated BV when changes are complete

All modifications follow the same conversational pattern: you describe what you want changed, the agent proposes a plan with the specific YAML edits, you approve, and the agent executes. After any change, you can ask the agent to re-run validation to confirm the BV is in a healthy state.