AI Search

Enable intelligent semantic search across your data using AI embeddings

What is AI Search?

AI Search transforms how you find information by understanding the meaning and context of your queries, not just matching keywords. Instead of searching for exact text matches, AI Search uses embeddings to find semantically similar content, enabling more intuitive and capable search experiences.

Snowflake Cortex Requirement: AI Search is exclusively powered by Snowflake Cortex. You must have a Snowflake Cortex provider configured with embedding services. OpenAI and Gemini providers cannot be used for AI Search functionality.

Prerequisites: AI Search requires a Snowflake Cortex embedding service. See Snowflake Cortex Setup for configuration instructions.

Table Type Requirement: AI Search works exclusively with Standard Snowflake tables. Other table types are not supported for search functionality.

How AI Search Works

AI Search uses machine learning embeddings to understand the semantic meaning of text content:

1. Content Processing

   Your text data is converted into high-dimensional vectors (embeddings) that represent semantic meaning

2. Query Understanding

   When you search, your query is also converted into an embedding using the same model

3. Similarity Matching

   The system finds content with embeddings most similar to your query, regardless of exact word matches

4. Results Ranking

   Results are ranked by semantic similarity and relevance to your query

Setting Up AI Search

Prerequisites

Before configuring AI Search, ensure you have:

AI Infrastructure

Snowflake Cortex Provider: Must be configured with embedding services

Standard Tables: Data stored in Standard Snowflake tables

Text Content: Fields containing searchable text data

Permissions: Appropriate access to configure search

Data Preparation

Clean Data: Well-formatted text content

Unique Identifiers: Primary keys for each record

Searchable Fields: Identified fields for search indexing

Reasonable Size: Tables sized appropriately for search performance

Configuration Process

1. Navigate to AI Search

   In the Intelligence tab, select AI Search from the menu

2. Add Search Table

   Click ”+ Search Table” to configure a new searchable table

3. Select Table

   Field to Search: Choose the primary field containing searchable text

   Standard Table: Select a Standard Snowflake table (other types not supported)

   Unique Identifier: Specify the field that uniquely identifies each record

4. Configure Search Parameters

   Attribute Fields: Select additional fields to include in search results

   Service: Choose your configured Snowflake Cortex embedding service

   Archival Service: Select Snowflake Cortex service to use for search queries

   Target Lag: Set how often to update search index (1 = daily updates)

5. Create Search Configuration

   Click “Create” to set up the search index

   Initial indexing may take time depending on data volume

Using AI Search

Search Interface

Once configured, AI Search provides an intuitive search experience:

Basic Search

Natural Language Queries: Ask questions in plain language

Semantic Understanding: Find content based on meaning, not just keywords

Contextual Results: Get results that understand the context of your query

Example Queries:

• “Issues with payment processing”
• “Customer complaints about delivery”
• “Technical problems with login”

Search Results

AI Search returns enriched results with:

Result Content

Matched Text: The content that matched your query

Similarity Score: How closely the content matches your query

Attribute Fields: Additional data fields you configured

Record Links: Direct links to full records

Search Insights

Query Understanding: How the AI interpreted your search

Match Reasoning: Why specific results were returned

Relevance Ranking: How results are ordered by relevance

Alternative Suggestions: Related searches you might try

AI Search in Automation

AI Search Records Action

Use AI Search in your automation workflows with the AI Search Records action:

Configuration

Purpose: Automatically find relevant records based on trigger content

Setup:

• Select your configured AI Search table
• Choose Snowflake Cortex embedding service for queries
• Set result limits (1-100 records)
• Configure attribute fields to return

Query Description: Provide guidance on what to search for (e.g., “Search for tickets by issue type”)

Best Practices:

• Use clear, specific search descriptions
• Set appropriate result limits
• Choose relevant attribute fields
• Test search queries before deployment

Use Cases

Customer Support: Find similar support tickets for faster resolution

Knowledge Base: Locate relevant documentation and procedures

Content Discovery: Find related content and resources

Duplicate Detection: Identify potentially duplicate records

Context Gathering: Collect relevant information for decision-making

Example Automation:

Email Received → AI Search Records (Knowledge Base)
  → Find: Articles related to email content
  → Return: Title, URL, Category, Last Updated
  → Send Email (with relevant articles)

Integration with Agents

AI Search powers conversational agents by enabling them to find relevant information:

Agent Tools

Knowledge Access: Agents can search your knowledge base

Context Gathering: Find relevant information for responses

Dynamic Responses: Provide current, accurate information

Self-Service: Enable users to find answers independently

Configuration

Search Tool: Configure AI Search as an agent tool

Result Limits: Set appropriate limits for agent responses

Attribute Fields: Choose fields for agent context

Filtering: Configure search filters for relevant results

Performance Optimization

Indexing Strategy

Content Optimization

Text Quality: Ensure high-quality, well-formatted text content

Content Length: Optimal content length for embedding models (512-1024 tokens)

Language Consistency: Use consistent language and terminology

Content Structure: Organize content logically for better understanding

Search Performance

Query Optimization

Query Clarity: Use clear, specific queries for better results

Result Limits: Set appropriate limits (10-50 results typically)

Caching: Cache frequent queries for faster responses

Batch Queries: Process multiple queries together when possible

Data Optimization

Table Size: Optimize table size for search performance

Field Selection: Index only necessary fields

Data Quality: Maintain high-quality, relevant content

Cleanup: Remove outdated or irrelevant content regularly

Best Practices

Content Preparation

1. Data Quality

   Clean Text: Remove formatting artifacts and ensure readable content

   Consistent Format: Use consistent formatting across all content

   Relevant Content: Include only content that should be searchable

   Complete Information: Ensure content provides complete context

2. Structure Optimization

   Logical Organization: Structure content in logical, searchable chunks

   Appropriate Length: Keep content chunks at optimal length for embeddings

   Clear Language: Use clear, professional language

   Avoid Duplication: Remove or consolidate duplicate content

Search Configuration

Embedding Selection

Model Choice: Use high-quality embedding models for best results

Consistency: Use the same embedding model throughout your system

Performance: Balance model quality with performance requirements

Cost Management: Consider embedding costs for large datasets

Attribute Configuration

Relevant Fields: Select fields that provide useful context

Field Limits: Don’t overwhelm users with too many fields

Data Types: Ensure attribute fields contain meaningful data

Performance Impact: Consider performance impact of many attributes

User Experience

Search Interface

Intuitive Design: Make search interface easy to use

Clear Instructions: Provide guidance on effective searching

Result Presentation: Present results in a clear, useful format

Feedback Mechanism: Allow users to provide feedback on results

Troubleshooting

Search Setup Issues

Symptoms: Cannot configure AI Search on a table

Common Causes:

• Table is not a Standard Snowflake table
• Snowflake Cortex embedding service not configured
• Insufficient permissions
• Invalid field selections

Solutions:

1. Verify table is a Standard Snowflake table
2. Ensure Snowflake Cortex embedding service is properly configured
3. Check permissions for search configuration
4. Review field selections and requirements

Poor Search Results

Symptoms: Search returns irrelevant or poor-quality results

Common Causes:

• Poor content quality
• Inappropriate embedding model
• Incorrect field configuration
• Insufficient content volume

Solutions:

1. Improve content quality and formatting
2. Try different embedding models
3. Review and adjust field configurations
4. Ensure sufficient content volume for training

Performance Issues

Symptoms: Slow search responses or timeouts

Common Causes:

• Large dataset size
• Inefficient queries
• Outdated index
• Resource constraints

Solutions:

1. Optimize dataset size and content
2. Improve query efficiency
3. Update search index regularly
4. Consider resource scaling

Advanced Features

Custom Search Filters

Attribute Filtering

Field-Based Filters: Filter results by specific field values

Date Ranges: Filter by date ranges for time-sensitive content

Category Filters: Filter by content categories or types

Custom Criteria: Create custom filtering logic

Dynamic Filtering

Context-Aware: Filters that adapt to user context

Role-Based: Different filters for different user roles

Workflow Integration: Filters that integrate with business workflows

Automated Filters: Filters applied automatically based on conditions

Analytics and Monitoring

Search Analytics

Query Patterns: Analyze common search patterns

Result Quality: Monitor result relevance and quality

User Behavior: Track how users interact with search

Performance Metrics: Monitor search performance over time

Integration Examples

Support Ticket Resolution

Support Ticket Created → AI Search Records (Knowledge Base)
  → Query: Ticket description and category
  → Results: Related articles and solutions
  → Update Record (add suggested solutions)
  → Notify Agent (with relevant resources)

Document Classification

Document Uploaded → AI Search Records (Document Categories)
  → Query: Document content summary
  → Results: Similar documents and categories
  → AI Classification (based on similar documents)
  → Update Record (assign category)

Customer Service Enhancement

Customer Email → AI Search Records (Previous Interactions)
  → Query: Customer issue description
  → Results: Similar past issues and resolutions
  → Generate Response (informed by past solutions)
  → Send Email (with contextual response)

Next Steps

With AI Search configured:

Build Agents

Create agents that can search your knowledge base

Use in Automation

Add AI Search to your automation workflows

Optimize Performance

Monitor and optimize search performance

Expand Coverage

Add more tables and content to your search system

---

AI Search transforms how you find and use information by understanding meaning and context. With proper configuration, it becomes a valuable tool for knowledge discovery and workflow enhancement.