Data exploration

Data exploration provides intelligent search and question answering across multimodal files in MOI. It can retrieve information from multiple documents and tables at once and generate summaries. Use it to locate information, review content, obtain answers, or analyze large collections of different file types.

Intelligent search and conversation

After entering a workspace, open Data exploration to access the search interface. Select Select files/tables in the upper-right corner to add search targets. A search can span multiple files and modalities, and multi-turn conversations retain context.

Note

Only files whose embeddings are complete and whose status is ready participate in retrieval. Disabled segments are excluded from recall.

If selected targets include tables, entering a question triggers a structured query (NL2SQL).

Configurable NL2SQL enhancements

To convert a natural-language question to correct SQL, provide the model with sufficient business context. Open the configuration page by selecting Settings in the upper-right corner of the data-exploration page.

Glossary

Define business terms and metric definitions. Add domain-specific terms and metric-calculation rules to help MOI understand the intended meaning of analytical questions.

Examples:

  • GMV: the sum of price × quantity for paid orders.

  • Valid order: an order whose status is paid, shipped, or completed.

Synonyms

Map multiple names to the same field. Configure common terms and their synonyms to help MOI understand analytical questions. Each canonical term can be bound to a specific table field.

Examples:

  • Sales ≈ revenue / turnover / GMV

  • User ≈ customer / member

Business logic

Describe complex rules and set them for system evaluation or global application. Add business definitions, specialized terms, and metric-calculation rules to help MOI understand business context in natural-language questions.

Examples:

  • “Recent” = the last seven days.

  • Exclude test_user_001 and test_user_002 globally.

Optimization case management

Optimization case management provides wildcard and case-library management. Wildcards work with the case library: add wildcards before adding cases.

Wildcards abstract variable parameters to make cases reusable.

Examples:

  • {city} = Beijing, Shanghai, Guangzhou, Shenzhen

  • {product_line} = phones, computers, wearable devices

The case library maps questions to standard SQL to guide SQL generation directly.

Example:

  • Question: Query sales for {product_line} in {city}.

  • SQL:

    SELECT SUM(total_amount)
FROM sales_orders
WHERE city = '${city}' AND product_line = '${product_line}';
Last updated on