Aiimi Insight Engine
User GuidesAiimi
  • Introducing Aiimi Insight Engine
  • Architecture
    • Overview and Key Concepts
    • Hosting Options
    • Architecture and How It Works
      • Agent Servers
        • Security Agent
        • Source Agent
        • Content Agent
        • Enrichment Agent
        • Job Agent
        • OCR Agent
        • Migration Agent
        • Tika Agent
      • Repository
        • Data Node
        • Proxy Node
        • Kibana Node
      • Gateway and User Interface
      • Document and Data Sources
    • Deployment Options
    • Security
      • User Security
      • Data and Document Security
      • Source System Security
      • Firewalling
      • Agent Servers
      • Repository
      • Gateway (Web Server)
      • Tools & Utilities
  • Installation
    • Elastic and Kibana Install (Windows)
    • Aiimi Insight Engine Installation (Windows)
      • Installation Security
      • Certificates in a Key Vault
      • SAR Configuration
      • CSOM Bridge Set Up
      • AI Studio
    • AI Services
      • Prerequisites
      • AI Enrichment Service
        • Installation and Setup
        • Enabling Enrichment Steps
        • Using AI Enrichment Steps
        • Performance and Concurrency
      • AI Model Service
        • Installation and Setup
        • Enabling Providers
        • Private Generative AI
        • Azure Open AI
        • Enabling AI History
        • HTML Cleaner Service
      • Configuration of Logging
      • Offline Set-up of Models
      • Using SSL
      • Running as a Service (Windows)
      • Using GPUs
      • AI and Semantic Search Set Up
        • Open & Closed Book AI
        • Semantic Search
          • Vectors for Semantic Search
          • Source Configuration
          • Sentence Transformer Models
          • Enrichment
          • Kibana
          • Final Search Flow
    • Email Threading Upgrade
  • Run Books
    • SharePoint Online Connector
      • Migrating ACS to Azure AD with Sites.Read.All
      • Migrating ACS to Azure AD with Sites.FullControl.All
  • Control Hub
    • Agents
      • Configurations
        • Config Management
        • Security Configurations
          • Security - General
          • Security - Source
            • Active Directory
            • Atlassian
            • Azure Active Directory
            • Builtin Security
            • Miro Security
            • Google Directory
            • SharePoint Security
            • Slack Security
          • Security - Sync
          • Security - Agents
          • Security - Scheduling
        • Source Configurations
          • Source - General
          • Source - Source
            • Alfresco Kafka
            • Azure Blob Storage
            • BBC Monitoring
            • Big Query Cataloguer
            • BIM360
            • CSV Data Loader
            • Confluence
            • Content Server
            • Data File Cataloguer
            • Document Store
            • DocuSign
            • Dropbox
            • Exchange 365
            • Filesystem
            • Google Bucket
            • Google Drive
            • Google Vault
            • Jira
            • JSON Data Loader
            • Livelink
            • Microsoft Teams
            • Mimecast
            • Miro
            • ODBC Data Loader
            • PowerBi Cataloguer
            • Reuters Connect
            • ShareFile
            • SharePoint
              • Azure Portal and Azure AD Authentication
              • Sensitivity Labels
            • SharePoint Legacy
            • SQL Server Cataloguer
            • Slack
            • Versioned Document Store
            • Websites
            • XML Data Loader
          • Source - Crawl
          • Source - Agents
          • Source - Schedule
          • Source - Advanced
        • Enrichment Configurations
          • Creating a Pipeline
            • General
            • Steps
              • AccessMiner
              • AI Classification
              • Apply Sensitivity Label
              • Anonymiser
              • CAD Extractor
              • Checksum
              • Content Retrieval
              • Copy
              • Data Rule Processor
              • Delete
              • Direct Copy
              • Email Extractor
              • Entity Rule Processor
              • External Links
              • Geotag
              • Google NLP Extractor
              • Google Vision Extractor
              • Metrics Calculation
              • Microsoft Vision Extractor
              • OcrRest
              • Office Metadata
              • PCI Extractor
              • REST
              • Set Document Risk
              • Text Cleaner
              • Tika Text Extraction
              • Trie Entity Extractor
              • Update Metadata
            • Filters
            • Agents
            • Schedule
            • Advanced
        • OCR Engine
        • Job Configurations
          • General
          • Job
            • AutomatedSearchJob
            • Command Job
            • ElasticJob
            • Extended Metrics Job
            • File Extractor
            • GoogleVaultSAR
            • Google Drive Last Access Date
            • Nightly Events Processor Job
            • Notifications Processor Job
            • Portal Sync Job
            • Purge Job
            • SAR Archiving
            • Text Content Merge Job
          • Output
          • Agents
          • Scheduling
        • Migration Configuration
          • General
          • Filter
          • Metadata Mappings
          • Agents
          • Scheduling
          • Advanced
      • Stats
        • Data Views
    • Security
      • User Settings
      • Credentials
      • Authentication
      • Application Access
      • Auditing
      • Descriptor Groups
      • Uploads
    • Mappings
      • Entities
        • Manage Entity Groups
        • Create an Entity
        • Manage an Entity
      • Models
        • Create a New Model
        • Find a Model
        • Enable or Disable a Model
      • Vectors
      • Rank Features
    • Search Settings
      • Search Relevancy
        • Core Settings
        • Makers Algorithm
        • Filename Boost Layer
        • Minimum Matching Terms Filter
        • Field Boost
        • Modified Date Boosting
        • Hit Highlighting
        • Why My Search Matched
        • Data Search Strategy
      • Bulk Search
        • Managing a Bulk Search
      • Filtering
      • Search Performance
      • Related Results
      • Featured Links
    • AI Settings
      • Search Flows
        • Search Flow Types
        • General Configuration
        • Query and Prompt Classification
        • Search Steps
        • Smart Filtering
        • Model Steps
        • Result Templates
        • System Prompt
      • Tools
        • Concepts
        • Import OOTB Tools
        • Built In Functions and Tools
        • Create and Edit Tools
      • Classifications
        • Class
        • Class Rules
        • AI Classification
    • User Interface
      • Thumbnails
      • Code of Conduct
      • Visualisations
        • Related Result Connections Diagram
        • Event Timeline
        • Timeline Lens Activity Chart
        • Relationship Map
      • Map Lens
      • Theming
      • User Avatar
    • Global Settings
      • General
      • App Settings
      • Presets
      • Metrics
      • Viewer
      • SAR
        • Importing Data For A SAR
        • SAR Disclosure Document Storage
        • Getting SAR data from Google Vault
        • SAR Configuration Access
        • SAR File Status
      • Disclosure Portal
        • Disclosure Portal Set Up
        • SARs From The Portal
        • Email Delivery Settings
          • Delivery Settings
          • Brand Settings
          • Customise Emails
        • SMS Delivery Settings
        • Requestor Message Limit
        • Attachment Configuration
        • Password Configuration
        • File Scanner Configurator
      • Collections
      • Notifications
      • OData API
  • AI Studio
    • Classifications
      • Classifications
      • Classification Rules
    • Jobs
  • Labels
  • API Guides
    • Insight API Guide
      • Swagger Documentation
      • Trying Some Endpoints
      • Search Filter
      • Hits / Items
      • Inspecting REST Calls
    • Data Science API Guide
      • REST Interface
        • Login
        • Datasets
        • Fields
        • Field Statistics
        • Search
        • Scroll
        • Update
      • Python Library
      • Data Science API Wrapper
        • Login
        • Datasets
        • Fields
        • Field Statistics
        • Search
        • Scroll
        • Scroll Search
        • Update Single Document
        • Bulk Update
      • Search API Wrapper
        • Login
        • Privileged Access
        • Search
        • Collection
        • ChatBot Class
      • Admin API Wrapper
      • AI Model Server API Wrapper
      • Utilities
        • Query Builders
        • Azure Key Vault Wrapper
    • Creating a Native Enrichment Step
      • Creating an Enrichment Step
        • Creating the Core Classes
        • Extending our Enrichment Step
        • Adding a Configuration Template
        • Adding the Enrichment Step
        • Creating an Enrichment Pipeline
      • Other Tasks
        • Entities, Metadata and Data
        • Accessing the Repository
      • Example Code
      • Troubleshooting
    • Creating a Python Enrichment Step
      • Creating an Enrichment Step
        • Running the Example from Command Line
        • Running the Example
      • Creating Your Own Step
      • Adding or Changing Entities, Metadata
  • Whitepapers and Explainers
    • From a Billion To One – Mastering Relevancy
    • Methods for Text Summarization
      • Application
      • Technology Methods
      • Commercial Tools
      • Key Research Centres
      • Productionisation
      • Related Areas of Text Analytics
      • Conclusion
      • References
Powered by GitBook
On this page
  1. Control Hub
  2. AI Settings
  3. Search Flows

Search Steps

PreviousQuery and Prompt ClassificationNextSmart Filtering

Last updated 2 months ago

CtrlK
  • Search Types
  • Standard
  • Cosine Similarity
  • Enrichment for Cosine Similarity
  • Rank Features
  • Enrichment for Rank Features
  • Configuration

Search Steps abstract the searching process, you can define different steps to get help users get the right results.

You can use a query classification step to select the correct search step. Or, if you're using tool calling, you can have the LLM select the correct search step.

Search Types

Standard

This is the standard BM25 keyword search. This is a general purpose algorithm that's good when the priority is returning everything that matches the query terms.

This is typically good for compliance scenarios.


Cosine Similarity

This is a type of semantic search. It uses dense vector embeddings and cosine similarity to compare a search to a set of documents.

  • To scale cosine similarity a keyword search is performed first to gather relevant items. These items are then reranked using cosine similarity.

You can limit the number of items reranked by defining a bucket size. This also helps scale the cosine similarity as it limits how many results go through it.

  • We recommend defining a bucket size if a keyword search returns more than a few thousand items.

When using a bucket size, the rerank is performed on the best matches defined by the keyword search. Items not reranked have their score normalised so they appear after anything reranked.

Documents and data must be enriched with dense vectors to use cosine similarity.

Cosine Similarity

Enrichment for Cosine Similarity

There are 3 key steps for enriching with cosine similarity. See our guide on creating vectors for more information.

  1. Create one or more Vector mappings in the Control Hub. These dimensions map to the AI model you are using.

  2. Apply the vector to the sources you want to use in this search flow.

  3. Create an enrichment step using AI Enrichment Service to compute and apply dense vectors.

Sentence Transformers

Note that you must make sure the vectors match across enrichment and search steps. For example, if you have uses a QA model at enrichment time, and you have a QA search step (i.e. you are using the query classification ask a question model), then you must use the same vectors.


Rank Features

Rank Features are a type of sparse vector. They map features or terms to weights that reflect their relative importance.

At search it compares the vector of the users search to the vectors in the results. It use the weights in the users query against the weights in the results to determine relevance.

To use rank features you will need to enrich your documents and data with sparse vectors.

Rank Features

Enrichment for Rank Features

There are 3 key steps for enriching for Rank Features. See our guide on creating rank features for more information.

  1. Create one or more Rank Feature mapping in the Control Hub.

  2. Apply the Rank Feature to the sources you want to use in this search flow.

  3. Create an enrichment step using AI Enrichment Service to compute and apply sparse vector.

Configuration

General configuration includes:

  • Step Name - The name of the step. If you have used query classification, this should be the same as one of the classification model labels.

  • Search Type - The type for the search step.

  • AI Model Service - The model service that is running the vectorizer provider.

  • AI Model Provider - The provider to use. You will only shown vectorizer providers.

  • Model - The model to use. Make sure you select the right model and that it matches the model uses at enrichment time.