Search AI

Search AI is an advanced RAG-based search solution that enables organizations to search across large datasets conversationally. It combines LLM capabilities with intelligent retrieval to generate accurate, context-aware answers.

Answers in Search AI are specific pieces of information extracted or generated in response to user queries. Unlike traditional search results that present ranked lists of documents, answers provide precise, tailored information directly addressing the user’s question.

This guide covers architecture, key concepts, the answer generation pipeline, and the complete setup workflow for configuring Search AI.

Architecture

┌────────────────────────────────────────────────────────────┐ │ User Query │ │ “What’s the return policy for electronics?” │ └──────────────────────────────┬─────────────────────────────┘ │ │ │ ▼ ┌────────────────────────────────────────────────────────────┐ │ Query Processing │ │ Query understanding → Intent detection → Query expansion │ └───────────────────────────────┬────────────────────────────┘ │ │ │ ▼ ┌────────────────────────────────────────────────────────────┐ │ Retrieval │ │ Vector search → Keyword search → Hybrid ranking │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │ │ │ Chunk │ │ Chunk │ │ Chunk │ │ Chunk │ │ │ │ 1 │ │ 2 │ │ 3 │ │ 4 │ │ │ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │ └──────────────────────────────┬─────────────────────────────┘ │ │ │ ▼ ┌────────────────────────────────────────────────────────────┐ │ Answer Generation │ │LLM synthesizes answer from retrieved chunks with citations │ └──────────────────────────────┬─────────────────────────────┘ │ │ │ ▼ ┌────────────────────────────────────────────────────────────┐ │ Response │ │ “Electronics can be returned within 30 days of purchase” │ │ [Source: Return Policy Document, Section 2.3] │ └────────────────────────────────────────────────────────────┘

Key Terminology

Term	Description
RAG (Retrieval Augmented Generation)	A method that extracts data from fragmented, unstructured content and formulates responses based on retrieved information. RAG retrieves both content and context from a dataset, significantly enhancing the precision and relevance of generated answers.
Chunking	The process of segmenting large content units into smaller, searchable segments. Search AI offers multiple chunking strategies based on content format.
Embeddings	Multi-dimensional vector representations of chunks, stored in a vector database. Different embedding models can be used for generating these vectors, enabling semantic similarity comparison.
Tokens	The smallest unit of data in computing terminology. A token represents a group of characters. Token Estimation: ~1 token ≈ 4 characters in English

Answer Generation Pipeline

The answer generation process consists of five sequential steps:

Step	Process	Description
1	Content Ingestion	Importing source documents that will be used to generate answers
2	Chunking	Breaking down source documents into smaller, meaningful units called chunks
3	Vector Embedding Generation	Converting chunks into multi-dimensional vectors for storage and comparison
4	Chunk Retrieval	Selecting the most relevant chunks based on similarity to the user query
5	Answer Generation	Generating a response to the user query using the retrieved chunks

Pipeline Flow:

Content Ingestion → Chunking → Vector Embeddings → Chunk Retrieval → Answer Generation

Setup Workflow

The Search AI setup consists of three main stages:

Stage	Purpose	Key Activities
Ingestion	Build unified knowledge base	Add content sources (websites, documents, connectors)
Enhancement	Refine and optimize content	Configure extraction, process via Workbench, set up vector generation
Retrieval	Configure answer delivery	Set retrieval strategy, query processing, answer generation

Stage 1: Content Ingestion

Integrate and index content from diverse data sources to build a unified knowledge base.

Content Source Types

Source Type	Description	Use Case
Web Crawl	Index content from websites	Public websites, knowledge portals
Directory	Upload and index documents	PDFs, Word docs, presentations
Connectors	Connect third-party applications	SharePoint, Confluence, ServiceNow, Google Drive

Supported Connectors

Search AI provides 100+ connectors including: SharePoint, Confluence, ServiceNow, OneDrive, Google Drive, Slack, Teams, Salesforce, Jira, HubSpot, Zendesk, and more.

Stage 2: Content Enhancement

Refine and enrich ingested content to improve answer quality.

Chunking Strategies

Strategy	Method	Best For
Text-based	Token-based segmentation; fixed number of consecutive tokens per chunk	General text content
Rule-based	Header-based segmentation; content between headers treated as chunks	Structured documents with clear headings

Workbench Processing

The Workbench tool enables content transformation through various stages:

Stage Type	Purpose
Field Mapping	Add, update, or delete fields
Custom Script	Apply JavaScript transformations
Exclude Documents	Filter content before indexing
API Stage	Enrich content via external APIs
LLM Stage	Refine content using LLM processing

Vector Configuration

Configure vector generation by selecting:

Configuration	Description
Embedding Model	Choose from supported models for vector generation
Fields for Vectorization	Specify which content parts (title, body, metadata) to use for embeddings

Stage 3: Retrieval Configuration

Set up retrieval and answer generation strategies for optimal results.

Retrieval Strategies

Strategy	Description	Best For
Vector Retrieval	Finds vectors most similar to the query vector	Semantic similarity, contextual queries
Hybrid Retrieval	Combines keyword-based and vector retrieval techniques	Balanced precision and recall

Query Processing (Agentic RAG)

Leverages LLM to enhance retrieval:

Capability	Description
Context Understanding	Interprets user intent from query context
Key Term Identification	Extracts important terms within queries
Noise Removal	Filters irrelevant content from queries
Accuracy Improvement	Enhances overall retrieval precision

Answer Generation Types

Type	Description
Extractive	Answers directly from content
Generative	Summarized and rephrased responses using LLM

Integration with Automation AI

Configure Search AI as a response method within AI Agents.

Answers Configuration

Navigate to: App Settings > App Profile > Enable Answers Feature

Intent Identification Priority Options

Option	Behavior
Automation first, Search AI as Fallback	Prioritizes automation framework; uses Search AI if intent identification fails
Search AI first, Automation as Fallback	Uses Search AI first; falls back to automation if no satisfactory match found

Additional Settings

Setting	Description
Use Search AI for Unrecognized Inputs During Dialogs	Passes unidentified user inputs to Search AI during Dialog Tasks, allowing knowledge base access mid-dialog

Setup Checklist

Step	Action	Location
1	Add content sources	Content > Sources
2	Configure extraction strategy	Index > Extraction
3	Process content via Workbench	Index > Workbench
4	Review chunks	Index > Content Browser
5	Configure vector settings	Index > Vector Configuration
6	Set retrieval strategy	Configuration > Retrieval Strategies
7	Configure answer generation	Configuration > Answer Generation
8	Enable Answers in App Profile	App Settings > App Profile
9	Test and debug	Configuration > Testing and Debugging

Key Capabilities Summary

Capability	Description
Extensive Data Connectivity	100+ connectors to enterprise content repositories
Enterprise Application Integration	Real-time access to business data from Salesforce, ServiceNow, Jira, etc.
Advanced Hybrid Search	Text and multi-vector weighted search with customizable pipelines
Enterprise Security Compliance	Integration with existing access control mechanisms
Contextual Intelligence	Dynamic enterprise and user context for personalized responses