Skip to Content
FeaturesIntent Detection

Intent Detection

Intent Detection is an intelligent system that automatically determines when DataQI assistants should query databases, use specific tools, or access different data sources based on user questions.

How Intent Detection Works

When you ask a question, DataQI assistants with Intent Detection:

  1. Analyze your question for data-related intent
  2. Evaluate available data sources to determine relevance
  3. Automatically decide whether to query SQL databases, search documents, or use other tools
  4. Execute the appropriate action without requiring explicit instructions

Currently Available

SQL Database Queries

Intent Detection automatically determines when to query connected SQL databases:

  • Data requests: “Show me sales for Q3”, “What’s our current inventory?”
  • Analytical questions: “Which products are performing best?”, “How are we trending?”
  • Real-time information: “What’s the status of…”, “How many users are currently…”
  • Comparative queries: “Compare this month to last month”, “Which region has the highest…”

Learn more about SQL Connections and how Intent Detection works with databases.

Configuration Tips

Data Source Descriptions

The more descriptive your data source descriptions, the better Intent Detection works:

Good Example:

“Sales database containing customer orders, product information, and revenue data from 2020-present”

Poor Example:

“Database”

Assistant Instructions

Provide clear context about when to use data sources:

Good Example:

“Use the sales database when users ask about revenue, orders, customers, or product performance”

Poor Example:

“This assistant can access data”

Future Capabilities

Intent Detection is being expanded to work with additional tools and data sources beyond SQL databases.

Best Practices

  • Use descriptive data source names and descriptions
  • Provide clear assistant instructions about when to use different tools
  • Ask specific questions that clearly indicate your intent
  • Review assistant responses to understand how intent was interpreted
Last updated on
Hallucination GuardrailSource Citations