The developers of the Real AI SA semantic analyzer have shared project updates after two years of development. The team has invested 8 ₽ million into the project, which aims to parse text into a graph format without relying on large language models (LLMs). The system currently contains approximately 5,000 concepts in its knowledge base.
Real AI SA is designed for tasks requiring 100% result repeatability and data privacy, as it can operate on hardware with as little as 4 GB of RAM. The analyzer processes text by categorizing elements into actions, properties, and objects. It currently supports only Russian-language text and requires manual input for domain-specific terminology.
The team successfully implemented a pilot project for an electronic document management system, where the analyzer reduced the time required to process orders from 7–10 minutes to 15 seconds. Despite this performance, the developers noted challenges in scaling the solution due to organizational resistance and the need for high document volumes.
The project is currently shifting focus toward maintaining document archives for small and medium-sized businesses. The developers emphasize that their approach avoids the rigid top-down ontology modeling that hindered previous projects like Compreno, instead utilizing a more flexible structure to manage complexity.
