AI Capabilities of IntelligenceBank

IntelligenceBank aims to ensure that Artificial Intelligence is incorporated responsibly into the overall platform, with efficiency gains, security and accuracy in focus as key pillars with every initiative.

Currently, Intelligencebank provides AI-powered information in two main areas: Asset Intelligence within the Digital Asset Management library and the Marketing Compliance module.

Asset Intelligence

This optional capability assists with tagging and labelling assets either upon initial upload or retrospectively. It includes the ability to tag imagery, video, and audio. Image tagging also allows for Facial Recognition, and Video/Audio includes Closed Captioning.

For Asset Intelligence, IntelligenceBank opted for Azure as the best-in-class based on a market analysis during development. The available data centers are Australia East, Canada Central, East US, and West Europe, which should match your IntelligenceBank AWS location.

Some key points to note are that this is a pre-trained model offering from Azure, and Azure is continuously improving its service independently. IntelligenceBank’s data is not used for training these models, and it cannot currently offer custom training.

Image Tag Categories

• General Keywords: Automatically detected from recognizable objects, living beings, scenery, and actions.
• Objects: Detected from objects or living things identified in the image.
• Brands: Automatically detected from commercial brands in the image.
• Landmarks: Automatically detected from identified landmarks in the image.

Process

• Image is instantly sent to Azure when uploading or editing
• Image Tagging returns content tags for thousands of recognizable objects, living beings, scenery, and actions.
• Dimensions of the image must be greater than 50 x 50 pixels.

Video/Audio Tag Categories

• Audiovisual Keywords: Auto-tags for Audiovisual Keywords are detected from insights on the different keywords discussed in media files. These keywords are detected from the speech and visual text and automatically tagged against the asset.
• Topics: Auto-tags for Topics are automatically inferred, based on various keywords from the speech and visual text.
• Objects: Auto-tags for Objects are automatically detected from visual objects and actions displayed.
• Locations: Auto-tags for Locations are automatically detected from the speech and visual text.
• Brands: Auto-tags for Brands are automatically detected from the speech and visual text.

Process/Restrictions

• Videos and audio files are post-processed; once saved, sent to Azure for tagging.

File Limitations

• Supported File Formats.
• File Size Upload limit optimized to 10GB (contact account manager for an increase).
• File Preview/Playback maximum 10GB, File Length maximum 4hr.

Closed Captions

• Sent post-processing to Azure, returns an SRT file.
• Users can upload their own SRT file to replace auto-generated captions.

Language Support

• Closed captioning supports a single language.
• Languages supported: https://learn.microsoft.com/en-us/azure/azure-video-indexer/language-support

Facial Recognition on Images

Identification

• Recognizes faces in images; creates a groupFace for the same person.
• Tagging a face with a name extends the tag to all associated faces/resources.
• If a face cannot be matched with any existing known face, it appears as Unknown.
  • To add a name, select from the drop-down menu or type a new name and select Create.
• If a Face is tagged incorrectly, there are three avenues:
• Rename Known Faces - Rename known faces directly from the Training Center.
• Detach Known Faces—Detach a face to remove the associated name and revert it to Unknown. This is helpful for correcting erroneous associations.
• Delete Known Faces—Permanently remove the name association from all images on the platform. This is useful for correcting widespread misidentifications.

Facial Attributes

• There is no emotion recognition or other facial attributes like age.

Privacy 

• Azure does not store faces; they are deleted after processing. We store a value called a persisted ID, which is the face attribute. This is used to match new faces. 

Ask AI: Advanced RAG Search Capabilities (Beta)

This optional feature is a powerful Retrieval-Augmented Generation (RAG) search capability that:

• Intelligently searches across your document corpus to find relevant information
• Generates concise, contextual summaries of complex content
• Provides references to source materials, ensuring transparency and traceability
• Enables users to quickly access and understand key information without reviewing entire documents

This advanced suite of AI capabilities, protected by our comprehensive guardrails system, allows IntelligenceBank to provide robust, adaptable, and secure document analysis tailored to the unique needs of each client.

The secure models are hosted in the same geographic region where your data is stored. If additional advanced rules are required that are hosted out of region, customers must approve first the usage of these models, noting they are routed via our LLM-gateway and guardrails platform for consistent security enforcement.

Marketing Compliance

The IntelligenceBank Marketing Compliance module incorporates a suite of advanced AI technologies designed to enhance content analysis, risk identification, and compliance management. 

Below is a detailed overview of the AI functionalities integrated into the Marketing Compliance module:

Agentic Compliance and Brand Rules

The platform leverages secure, state-of-the-art large language models to detect text-based risks in documents, including validating the presence or absence of associated disclaimers or proprietary statements as required by business rules.. These rules are highly customizable and can be configured based on:

• Prompts.
• Predefined rule lists.
• Source-of-truth documents such as terms and conditions.

Natural Language Processing (NLP)

NLP capabilities within the platform extend beyond basic content extraction to include:

• Spelling and Grammar Checks: Ensures content adheres to language conventions.
• Named Entity Recognition (NER): Identifies and categorizes entities (e.g., names of individuals, products, or organizations) for rule-based risk assessments.
• Other intelligent rules that enhance document compliance and quality.

Sentence Embedding and Similarity Rules

The platform leverages secure, state-of-the-art large language models to convert sentences into data points for comparison. This enables:

• Identification of similar or dissimilar sentences based on business rules.
• Enhanced contextual analysis for nuanced risk detection.

Named Entity Recognition (NER)

NER is used to identify and classify entities within content, such as:

• Determining whether a name refers to a person, product, or company.
• Facilitating rules that validate entity qualifications or compliance standards.

Computer Vision

The platform integrates Google Vision for diverse use cases, particularly during the content extraction phase. This ensures accurate identification of document components before applying risk rules.

Optical Character Recognition (OCR)

OCR functionality, supported by machine learning, ensures accurate text extraction even in challenging scenarios, such as:

• Obscured text due to images or stylized fonts.
• Predictive text recognition to handle incomplete or unclear content.

Image Comparison

The platform includes image recognition and comparison capabilities, often used for:

• Logo Detection: Identifying the correct or incorrect presence or absence of client logos and flagging related risks.
• Rule Application: Validating the presence or absence of associated disclaimers or proprietary statements as required by business rules.

Image Labelling

Using pre-trained or client-specific models, the platform assesses images in documents to flag potential risks. For example:

• Identifying imagery that may not comply with client-specific guidelines, such as avoiding images of children in certain marketing materials.

Data Security and Privacy

The platform ensures strict data security standards:

• No client data is sent to third-party models or LLMs for training purposes.
• Our AI-gateway and AI guardrails provide an additional layer of protection and control over all AI interactions.
• Agreements with third-party AI providers ensure that client data remains private and is used exclusively for the intended processing.

This advanced suite of AI capabilities, protected by our comprehensive guardrails system, allows IntelligenceBank to provide robust, adaptable, and secure document analysis tailored to the unique needs of each client.

The secure models are hosted in the same geographic region where your data is stored. If additional advanced rules are required that are hosted out of region, customers must approve first the usage of these models, noting they are managed within our LLM-gateway and guardrails platform for consistent security enforcement.