On February 11, 2025, Intecracy Group consortium hosted another Intecracy Solution Day in a mixed format. The main topic of the event was the integration of Large Language Models (LLMs) into real-world business processes of enterprises and organizations. The participants discussed practical aspects of artificial intelligence implementation, focusing on critical issues of security, access control, and corporate data confidentiality.
The event was led by IQusion IT LLC, an Intecracy Group member.
Architectural Challenge: How to Combine AI and Corporate Data
Modern businesses strive to leverage the capabilities of generative artificial intelligence to automate routine operations, retrieve information, and support decision-making. However, directly connecting public or even local large language models to corporate databases poses serious risks. The main challenge is that LLMs are prone to hallucinations and lack built-in tools to enforce corporate security policies and access control. That is why the central theme of this year's Intecracy Solution Day was artificial intelligence access to corporate data and functions through a strictly controlled architecture.
The event organizers emphasized that successful AI integration requires the creation of an intermediate protective layer. This layer must act as a mediator between the intelligent agent and the company's information systems. The Ukrainian-developed UnityBase platform, which allows building fast and secure web applications based on metadata, is considered the technological foundation for this approach. Due to its architecture, UnityBase is capable of providing tight control over exactly what data is passed to the AI model and what actions it can perform within the system on behalf of the user.
Controlled Access: Security Mechanisms by Serhii Balashuk
In his presentation, leading architect Serhii Balashuk detailed the technical tools for implementing this approach. He demonstrated how the UnityBase platform can be used to set up a secure gateway for interacting with large language models. The core principle is that artificial intelligence never gains direct access to the database or file storage.
"We must treat the LLM as an external operator delegated with specific tasks, but under the strict supervision of an access control system. Using the capabilities of UnityBase, we create a specialized API layer. The model communicates with the system exclusively through this interface, where every request undergoes rigorous validation. If the user who initiated the request to the AI does not have permission to view a specific document, the model physically cannot retrieve this document for analysis. This completely eliminates the risk of confidential information leaking through intelligent interfaces," noted Serhii Balashuk.
This approach also addresses the issue of executing actions. Instead of allowing the AI to independently make changes to registries or databases, UnityBase generates transactional proposals that require mandatory human approval or pass through predefined business rules. This ensures that no automated decision by the artificial intelligence leads to irreversible errors in critical business processes.
Strategic View on Process Transformation by Yurii Syvytskyi
Yurii Syvytskyi, co-founder of Intecracy Group consortium, dedicated his speech to the business context of implementing new technologies and the strategic trade-offs companies must face. He emphasized that artificial intelligence technologies are no longer just toys for IT departments but are becoming real tools for competitive advantage.
"Today, businesses face a choice: either ignore the development of LLMs and lose efficiency, or implement them chaotically, exposing themselves to colossal security risks. Our position is clear — integration must be systematic. The controlled architecture we offer based on UnityBase allows maintaining a balance between the speed of innovation and business security. We do not try to retrain the artificial intelligence or force it to work without errors. Instead, we build a reliable environment around it that simply will not allow the AI to make a critical mistake or exceed its permitted authority," explained Yurii Syvytskyi.
Yurii Syvytskyi also pointed out that using local (on-premise) language models is becoming increasingly relevant for large enterprises and government institutions. This allows for the complete localization of data processing within the organization's perimeter, which is crucial for compliance with legislation regarding personal data protection and state secrets.
Technological Stack and Implementation Prospects
During the practical session of Intecracy Solution Day, the speakers demonstrated several scenarios of user interaction with the corporate system using natural language. Through the integration of UnityBase with modern LLM models, users can generate complex analytical reports, search for relationships in documents, and automate the preparation of standard responses using familiar text commands.
An important aspect of the discussion was solution performance. Since UnityBase is developed using high-performance server-side JavaScript technologies and optimized for high-load operations, adding an intelligent layer does not lead to a significant slowdown of the system. This opens up wide opportunities for scaling solutions to thousands of concurrent users without requiring an extraordinary increase in hardware infrastructure costs.
The event concluded with an open discussion where participants discussed further steps in platform development and the possibilities of pilot implementations of the controlled AI architecture in various sectors of the economy. Intecracy Group continues to work on improving its products, helping businesses safely enter the era of artificial intelligence.