Cogniware, s.r.o, delivers innovative AI-based analytical products and elite professional services to help its clients identify potential safety threats and uncover hidden patterns of suspicious behavior and crime. Cogniware supports clients with an AI-powered data analysis platform called Argos, a security-rich data analytics solution for investigation, deep analysis, pattern search and more. The Argos platform was designed to efficiently collect, link and analyze various data formats, yet Cogniware saw an opportunity to further increase the platform’s usability.

By leveraging AI to handle more complex data interactions, Argos could be useful to sectors, such as law enforcement and commercial banking, where quick and accurate data analysis is vital. Therefore, Cogniware aimed to streamline the user experience and increase the platform’s overall efficiency. Making the system even more intuitive and user-friendly, so users can query data using natural language and tailor the platform to clients’ specific needs, reflecting Cogniware’s commitment to making Argos useful to all investigators.

Cogniware partnered with IBM Client Engineering to pilot the integration of IBM® watsonx.ai™ LLM models into the Argos platform. The team focused on enabling users to query data in natural language, with the system automatically generating analytical patterns. The pilot included people from Cogniware and IBM and ran for several months. The scope of the pilot included internal data from Cogniware, as well as external datasets, ensuring a robust evaluation of the system’s capabilities.

During the pilot, the key factor was using LLMs to translate free-text queries into structured analytical patterns, thereby minimizing the risk of model hallucinations, which could provide false or misleading results. According to Dominik Regner, Sales Manager at Cogniware, “For critical decisions, it is necessary to have precise results; there is no room for hallucinations or mistakes made by the LLM.” The pilot incorporated LLMs to summarize texts and extract entities, enhancing the overall efficiency of the platform. “We focused on two main goals—be able to process unstructured documents with LLM and add extracted knowledge into our graph and be able to efficiently query graph data with natural language—all in an intuitive and highly secured way. Our clients need to have absolute trust in the safety measures for their sensitive data. Results of our cooperation with [the] IBM team is a game-changer for our path to deliver self-service analytics software to business-oriented users,” said Martin Kovář, Product Manager at Cogniware.