Empowering Data Teams: Smarter AI Workflows with Hex & MotherDuck

The Key Insight: Context is King

LLMs don't know your business data. The most important thing for accurate AI analytics is building context:

1. Column descriptions in your database
2. Semantic models defining business metrics
3. Rules files with business logic and SQL styling
4. Endorsed assets marking trusted tables

Generating Column Descriptions with AI

MotherDuck's prompt() function generates descriptions directly in SQL. The workflow involves:

• Getting a table summary for AI context
• Generating AI descriptions focused on business purpose
• Applying descriptions to your schema with COMMENT statements

Pro tip: Tell the AI to focus on business purpose, not statistics. Avoid descriptions like "contains integers from 1-100"—that's already in the schema.

Read Scaling for Concurrent Users

MotherDuck's read scaling gives each Hex user their own DuckDB instance:

• Configure in Settings → Instance Size
• Choose fleet size (up to 16 ducklings)
• Eliminates noisy neighbor problems
• Pay only for what you use

Hex's Three Pillars for AI Workflows

Demo: From Question to Dashboard

User prompt: "Break down our marketing opportunities"

What happens automatically:

1. Fuzzy search finds related dashboards
2. Loads semantic model definitions
3. Runs 20+ iterative SQL queries
4. Creates charts, pivot tables, insights
5. Generates a complete notebook

Converting to production:

• Thread → Project (for data team review)
• Project → Dashboard (drag-and-drop app builder)
• Dashboard → Semantic Model (close the loop)

Context Studio: Monitoring & Improvement

Track how users interact with AI:

• Conversation volumes and top users
• Actual prompts being submitted
• Which queries hit semantic models vs raw tables
• Identify hallucinations and gaps

Rules files allow you to define business definitions (like what "Revenue" or "Active customer" means) and SQL style preferences in markdown format.

Best Practices