Reliable data here and now is already half the success in the business environment. At least, this is what most of our Customers and Partners say. That is why we are constantly improving the Knowledge On-Demand function – a tool for intelligent data search throughout corporate knowledge bases, implemented in the UniBot Personal Assistant & Corporate Chatbot. We introduce our own developments and implement the latest Artificial Intelligence models to constantly increase the relevance of the search. And in this material, we share the latest changes in AI-UniBot by Lizard Soft.

• The new content extraction mechanism

Among the main innovations is, first of all, the new mechanism for extracting content from documents. It now uses the latest Azure AI models for OCR (Optical Character Recognition) and Computer Vision (image and video recognition, classification, and analysis). These improvements are particularly relevant to algorithms for recognizing tabular data, graph data, and business process diagrams. This means that even complex tables or graphical representations of processes can now be automatically recognized and integrated into the system for further analysis and use.

• The new mechanism for semantic chunking of information

The new mechanism of semantic chunking of information allows working with large documents much more efficiently. Since each AI model has a limit on the size of information that can be worked with within the single request framework, uploading large documents becomes economically impractical. Instead, the new mechanism allows you to cut the document into parts (chunks) in such a way that each of them is searchable and contains a full context. This greatly improves the AI's ability to draw correct conclusions based on the information provided.

• The new format of the semantic index with a strict search mode and specific features of language understanding

The new format of the semantic index introduced in AI-UniBot now enables strict search mode and understands the language nuances of the indexed text. For example, if the text contains the word «individuals» and the User searches for the phrase «natural person», the system – realizing that the meaning of both phrases is common – could find it. This innovation allows to reduce the number of iterations required to form a response to the User's request. And this, in turn, reduces the waiting time and saves the Company's resources.

Updated features of AI-UniBot open up new possibilities for applying Knowledge On-Demand in various business cases. For example, in Companies where a large amount of data is stored in complex tables or diagrams (such as business process descriptions), the new mechanisms of extraction and semantic indexing would prove themselves best. Thanks to them, the necessary information would be provided as quickly and accurately as possible. Another example is Companies that have a significant number of documents on a similar topic with an inconspicuous difference. For example, technical specifications of products or contracts with various Clients. The new chunking mechanism performs most effectively with such documents.

In short, the latest updates in AI-UniBot have significantly increased the efficiency and accuracy of intelligent data retrieval. Thus, the Knowledge On-Demand tool has become even more useful for businesses. Thanks to the new capabilities of content extraction, semantic chunking, and indexing, Users get reliable information in response to their queries even faster, no matter how complex or large-scale the searched content is. This means that the right decisions would be made even faster, and this is already half the success, most of our Customers and Partners claim.