European Funds — Trilagi

TRILAGI SPÓŁKA Z OGRANICZONĄ ODPOWIEDZIALNOŚCIĄ is carrying out a project co-financed by the European Funds entitled “Cognitive RCA diagnostic system with adaptive contextual chunking and an agent architecture”. (Original title: „Kognitywny system diagnostyki RCA z adaptacyjną fragmentacją kontekstową i architekturą agentową”)

The aim of the project is to close two critical gaps in scientific knowledge that currently limit the effectiveness of automated diagnostic systems to just 29% (according to Roy et al., 2024). The first is the inadequacy of standard data chunking techniques for specialised technical information, which leads to a loss of semantic integrity in key diagnostic data. The second is the lack of effective multi-agent collaboration models for technical diagnostics, which prevents the different aspects of diagnostic analysis from being combined effectively.

Tasks planned in the project:

• Research on adaptive contextual chunking (MAFK) of heterogeneous data for Linux system diagnostics – R&D
• Research on a hierarchical agent architecture with context-driven communication (MHKA) for Linux system diagnostics – R&D
• Experimental development of a prototype integrating MAFK and MHKA – R&D

Project costs:

• Project staff
• Depreciation of equipment
• External services related to colocation, cloud and tool subscriptions, and access to LLMs

Target groups:

The project is aimed at companies and organisations with server infrastructure based on Linux systems, in particular system administration, DevOps, SRE and ITOps teams and providers of IT infrastructure maintenance services. The solution will be used by people responsible for incident analysis, diagnosing the causes of failures and ensuring the continuity of digital services. Indirect beneficiaries will be the customers and end users of these services, who will benefit from shorter diagnosis times and reduced downtime.

Project results:

As a result of the project, the Beneficiary will implement the Cognitive RCA Diagnostic System (KSD-RCA) – an AI platform based on the Adaptive Contextual Chunking Model (MAFK) and the Hierarchical Agent Communication Model (MHKA). A unique feature of the solution is that it uses existing data (logs, monitoring metrics, knowledge base, ticket content), so it does not require any additional services to be installed. The tool will increase the accuracy of RCA analyses from 29% to 55% and reduce the average failure diagnosis time from 120 min to ≤ 45 min. A chat interface will present a transparent reasoning path, making it easy to quickly understand the steps taken during RCA. As a result, using the tool will not require additional staff training. The solution will provide a significant technological advantage over competitors by shortening incident diagnosis time and significantly increasing the likelihood of correctly identifying incident causes.

Project value: PLN 3,452,554.76

European Funds contribution: PLN 2,604,817.29