Skip to main content

Review Relationships

After ERD generation, use the AI Agent Chat to review, validate, and refine detected relationships through natural language conversation.

Prerequisites

Complete the Generate ERD step to create the Entity Relationship Diagram.

Agentic Chat Review​

The AI Agent Chat remains available after ERD generation for interactive refinement:

What You Can Do​

  • Validate — Confirm AI-detected relationships
  • Challenge — Question uncertain detections
  • Modify — Request relationship changes
  • Add — Specify missing relationships
  • Remove — Delete incorrect relationships

Example Commands​

"Why did you choose CUSTOMER_ID as the FK
in TEMENOS_ACCOUNTS?"

"Add a relationship between CARDS and
FRAUD_ALERTS on CARD_NUMBER"

"Remove the relationship between
CAMPAIGNS and INTERACTIONS"

"The PRODUCT_ID relationship should be
to PRODUCTS, not CATEGORIES"

Review Workflow​

1. Review Generated Relationships​

Navigate through the RELATIONSHIPS tab to see all detected relationships:

Source TableSource ColumnTarget TableTarget ColumnConfidence
SALESFORCE_CAMPAIGNSPRODUCT_IDSALESFORCE_PRODUCTSPRODUCT_ID0.90
TEMENOS_ACCOUNTSCUSTOMER_IDTEMENOS_CUSTOMERSCUSTOMER_ID0.95
WORLDLINE_CARDSCUSTOMER_IDTEMENOS_CUSTOMERSCUSTOMER_ID0.88

2. Challenge Uncertain Relationships​

For relationships with lower confidence, ask the AI for justification:

You: "Why is CREATED_BY linked to CUSTOMERS?"

AI Agent:

"I detected this relationship based on:

  • Column name pattern suggests a user reference
  • Value distribution shows overlap with CUSTOMER_ID values (78% match)
  • However, confidence is medium (0.65) because CREATED_BY could also refer to internal user IDs

Would you like me to remove this relationship, or can you confirm whether CREATED_BY references customers or internal users?"

3. Modify as Needed​

Request specific changes:

You: "CREATED_BY references internal users, not customers. Remove that relationship."

AI Agent:

"Understood. I've removed the CREATED_BY → CUSTOMERS relationship. The ERD has been updated.

Is there a different table I should link CREATED_BY to, such as an INTERNAL_USERS or EMPLOYEES table?"

Confidence Scores Explained​

ConfidenceScore RangeMeaning
Certain0.95 - 1.00Strong statistical and semantic evidence
High0.80 - 0.94Good evidence, minor ambiguity
Medium0.60 - 0.79Some evidence, review recommended
Low< 0.60Weak evidence, manual verification needed
Focus Your Review

Start with Medium and Low confidence relationships. High confidence detections are usually accurate.

Common Review Scenarios​

Scenario 1: Missing Relationship​

You: "FRAUD_ALERTS should link to TRANSACTIONS via TRANSACTION_ID"

AI Agent:

"I'll add that relationship. Analyzing the columns...

  • FRAUD_ALERTS.TRANSACTION_ID exists (NUMBER)
  • WORLDLINE_TRANSACTIONS.TRANSACTION_ID exists (NUMBER, PK)
  • Value overlap: 100%

Relationship added: FRAUD_ALERTS.TRANSACTION_ID → WORLDLINE_TRANSACTIONS.TRANSACTION_ID

ERD updated."

Scenario 2: Incorrect Target Table​

You: "The PRODUCT_ID in CAMPAIGNS links to PRODUCTS, not PRODUCT_CATEGORIES"

AI Agent:

"I'll update the target table. Changing:

  • FROM: CAMPAIGNS.PRODUCT_ID → PRODUCT_CATEGORIES.CATEGORY_ID
  • TO: CAMPAIGNS.PRODUCT_ID → SALESFORCE_PRODUCTS.PRODUCT_ID

Relationship updated."

Scenario 3: Composite Key Clarification​

You: "TRANSACTION_DETAILS uses a composite key of (TRANSACTION_ID, LINE_NUMBER)"

AI Agent:

"I've updated TRANSACTION_DETAILS to use a composite primary key:

  • PK: (TRANSACTION_ID, LINE_NUMBER)
  • This replaces the previous single-column PK detection

The TRANSACTION_ID portion maintains its FK to TRANSACTIONS."

Audit Trail​

All review decisions are logged:

TimestampActionBeforeAfterUser
2026-01-08 14:32RemoveCREATED_BY→CUSTOMERS—hussnain@ekai.ai
2026-01-08 14:35Add—FRAUD_ALERTS→TRANSACTIONShussnain@ekai.ai
2026-01-08 14:38ModifyCAMPAIGNS→CATEGORIESCAMPAIGNS→PRODUCTShussnain@ekai.ai

Completing Review​

When satisfied with the ERD:

  1. Verify all relationships in the diagram
  2. Download DBML if needed for external tools
  3. Proceed to create a Semantic Model using this ERD
ERD Ready for Semantic Models

Your reviewed ERD can now be used to create Semantic Models with DBT projects, data catalogs, and more.

Next Steps​