This article will provide an overview to all things Atlas IKS, including IKS Set-up & Configuration as well as information on Knowledge Collections (KCs) and how to set them up. Please be aware that our IKS engagements revolve around specific AI Use Cases applicable to your firm. Information on Use Cases and related Documentation for AI engagement will be discussed in our workshops and provided via files - this information will not be contained within this article.
*See all Atlas AI & IKS articles here*
- Overview
- Prerequisites
- How to Access the IKS
- Knowledge Collections
- Managing the IKS
- Auditing & Feedback
- Example Use Cases
Overview
The Intelligent Knowledge Studio is an innovative software tool designed to empower authorized users to create specialized "knowledge collections" . These collections are curated sets of content grounded within an AI index, ensuring that the information is accurate, relevant, and authoritative.
Each user that has permissions to the Atlas AI Assistant will be able to interact with multiple knowledge collections. Each Knowledge Collection is a configuration of the AI with access to specific information sources, using a specific AI model. The model, settings and information can then be tailored to specific use cases to ensure optimized results.
This means you can set up Knowledge Collections that only have information you have carefully chosen to support a specific use case, using cheaper models for basic Q&A functionality, and more expensive models for more complex requirements.
Prerequisites
There are a number of prerequisites to manage and be aware of before you will be able to use and interact with Atlas AI.
- AI tool must be deployed and the Azure configuration and technical back-end deployment in the Atlas infrastructure must be completed via our Runbooks. IKS requires Atlas v6.0.
- AI tool is sat behind a feature flag which must be turned on by ClearPeople after the AI infrastructure has been deployed to Azure.
- Use Cases must be built and worked on to understand what you are wanting to achieve out of AI. This will help provide and plan for a roadmap of configuration of both AI, AI models, AI settings, and the various KCs you will want to test content against.
- An AI Pilot engagement is necessary to strategise your use of AI in the correct way. It's not a one-size fits all, and to really make the most out of AI we advise a smaller Pilot is taken forward and completed before being rolled out to end-users.
- Budget - unfortauntely AI is not free. Microsoft will charge you additional consumption fees for the extra AI search indexing and also housing and running the necessary AI infrastructure.
- Permissions need to be allocated for use. You can read in-depth information about Permissions, access and governance in this article.
Permissions
Please read the article linked to in the above bullet point around permissions for an in-depth overview and explanation of the permissions set-up for IKS and the different roles involved and what permissions each role provides.
Please note that to interact with the AI Assistant (chatbot), which is the way you interact with Knowledge Collections (the information and data stored within the IKS tool), the AI Assistant has it's own set of permissions, and although they are 'both sides of the same coin', they are by definition different features. Please read the Atlas AI Assistant Overview article as well as the Atlas AI Assistant Configuration article for more details on the AI Assistant.
AI Assistant Access
To interact with (speak to) KC content via AI, this needs to be done via the Atlas AI Assistant, and therefore any users who need to leverage AI needs to be part of the Atlas AI Assistant Users Azure Entra ID Security Group.
Accessing IKS and creating, managing or auditing KCs is managed via a different set of permissions.
How to access the IKS
You will need appropriate permissions in order to access the IKS (see previous linked article on permissions). Once you have permissions for your role in the AI engagement, you can find the IKS tool within the 'My Atlas' icon within the Atlas Global Navigation (Main Menu). It is stored within Advanced Tools.
If you have permissions to see IKS, it appears as shown below. If you do not see this IKS option, you'll need to review your permissions to access.
Knowledge Collections
The main asset or options within the wider IKC tool are called Knowledge Collections (known as KCs). They are essentially what the name suggests - collections of documents and other text-based content which hold all the information you want the Atlas AI to be able to use to answer questions on a specific topic. Currently this is limited to text-based content such as Office documents, PDFs, structured or unstructured text files, as well as information within SharePoint Lists etc.
In the below example you can see a list of KCs. The green dot on the left signifies the state of the search indexing against the content within the KC. The icon and the numbers on the right show how many articles are included in the specific KC.
Accessing Knowledge Collections
Each KC has it's own permissions, both to be able to use the KC via the Atlas AI Assistant, but also to both administrate and/or audit KCs. Depending on your permissions you may not be able to see any KCs, or you might see all KCs (if you are an administrator).
Use of each Knowledge Collection is subject to the permissions applied to the content, so users can only see and interact with Knowledge Collections they have been granted access to. Additionally, the AI checks user credentials before accessing information to use as part of a response. If a user does not have the necessary permissions to a certain document for example, the AI will not be able to retrieve that information for any responses so that user might get a different answer than another user.
Creating a Knowledge Collection
While we refer to this as "creating" a Knowledge Collection, the data you want to include needs to already exist in an Atlas document library. However you need to create the Knowledge Collection in IKS to give the AI access to the documentation.
- Navigate to Intelligent Knowledge Studio: Go to the administration dashboard.
- Create New Collection: Click on "+New" in the top-right
- Naming Conventions: Use clear and descriptive names for the Title field in your new KC so users know which KC they need very clearly.
- Configure Permissions: Set the audience for the collection. Only users in the collection audience can see and interact with it, all other users cannot. Owners and Auditors have other permissions related to IKS, but will not be able to interact with the KC through the AI Assistant. All 3x security fields are mandatory. If you need to leave it blank but are responsible for creating and managing the KC going forward, please put yourself and the project team, or one of our Atlas IKS Azure Entra ID Security groups, into these fields.
-
Model Settings: Provide context to the AI by setting instructions. For example, instruct the AI to be an expert assistant in Atlas architecture and technical questions. These instructions will have a considerable impact on the KC and how the AI Assistant interacts with users.
There is also the ability to select the AI GPT model to be used against this KC. We currently recommend 40 mini as it's the best value. You can read more about models here. Please note your firm can also create their own custom models in Azure OpenAI. Any custom models available in your tenant will appear here.
You will need to pick a model before moving onto the next section
-
Grounding Settings:
This is a really important section to understand.
-
- Strict Grounding: If set to Yes, the model will be strongly instructed to not generate any responses unless sourced in relevant information in the knowledge sources configured including references to verifiable sources. If set to No, the model will be open to generate responses that may or may not include information from verifiable sources. In summary, if you don't want the AI to try and make stuff up from what it understands purely from a model perspective, ensure strict grounding is on. It will be set to Yes by default. Grounding responses will ensure information is accurate and contextually relevant.
- Number of Search results used to respond: Represents the number of chunks/facts that will be retrieved as a result of a search operation from the grounded knowledge to be then filtered by relevancy. The higher your score here, the more information it will review in order to answer your question. Higher results will be more detailed, have a higher chance of accuracy but will be more expensive.
- Relevancy score for search results: Represents the minimum relevancy for each of the search results (chunks/facts) retrieved in order to consider them relevant enough for the model to use them as source of knowledge to generate a response. A higher relevancy score will ensure that only the best matching content is returned to you, but it may limit the responses you receive as the 'pool' of chunks/facts will be smaller. However having a relevancy score too low will 'wash-out' your responses with answers that are potentially not what you were expecting as the model is attempting to answer your questions base to the best of its ability based on the information it has at it's disposal. It is similar to Search Relevancy in terms of specific content receiving a higher rate of relevance to your search.
- Knowledge Sources: In this area you will be able to select the Atlas Workspaces you want to select data from for your KC
When you search for and find the correct Atlas Workspace, after you click it, you will see more options load, dynamically showing the lists and libraries available to choose from. You can select multiple options here.
After you have confirmed a selection, the screen will show the saved options. In the below example I have selected multiple lists and libraries from 3x different workspaces.
Apply KQL Query Filter: In this box you will be able to write queries in Keyword Query Language (SharePoint Search Language) to further refine and trim the results based on your requirements and use cases.
In the example below, I have input a query to say 'only show me .pdf documents' and you can see that after I have 'tested' the query, there will be 18 documents included in this knowledge collection.
If your query does not work or has not been written correctly you will see 0 results. This is not always essential, but this is where you further leverage the Atlas Taxonomy (metadata tags), so if you only want one type of document shown (such as RFPs or Project Plans) or if you just want information related to 1x Department, 1x Location, 1x Entity, 1x Subject or 1x Activity, or a combination, this can be a very powerful tool for telling the model explicitly to only use certain documents.
Using a Knowledge Collection
Once created and crawled, and appropriate permissions have been granted, a user will be able to choose which Knowledge Collection to interact with before they can ask a question to the AI Assistant. This is part of the reason good naming for your collections is important.
As a user, to understand how to query the AI please see this article: IKS - Interact with Knowledge Collections as a User
- Interacting with the AI: Users can ask questions, and the AI will respond based on the chosen Knowledge Collection.
- Security Trimming: The AI will only provide information if the user has access to the source documents. Permissions are synced every 30 minutes.
Managing the IKS
Managing the IKS tool is different to interacting with KCs through the AI Assistant, which we hope with the information provided so far in this article and the permissions article should now be well understood.
Managing the IKS falls into several categories based on roles and responsibilities. Please view the permissions article linked at the top of the page for more information on which Roles are centrally managed. This is different to allowing management of a KC.
Of course a key consideration of managing the IKS is actually managing the permissions, but we will not go through that in this article.
Those with the appropriate permissions will be able to Edit a KC, run a Full Sync or Incremental Sync, can Show details, and also Delete
There's also options along the top row to Reports and Export Feedback. These are discussed in the next section of this article below called Auditing. But first, we need to present how you manage the Technical Settings and Configurations of the AI & IKS settings in Azure.
Technical Configuration & AI Settings in Azure
Please view this article for in depth overview and explanation of all settings in Azure related to AI & IKS. Please note this is a technical subject with technical and AI-Based language used throughout.
Auditing & Feedback
Please view this article for an overview and insight into IKS & AI 'Auditing'
Example Use Cases
- Atlas Technical Documentation: A collection that includes both internal and Zendesk documentation, accessible to the internal team for technical Q&A.
Comments
0 comments
Please sign in to leave a comment.