More | OLMapp

Ontological Language Model (OLM) is a new approach to how machines understand, reason, and control what they communicate.

It is a new language-model/architecture designed to operate independently or alongside LLMs. Instead of relying exclusively on statistical prediction, OLM introduces a deterministic layer built around structured meaning.

The benefits of OLM are grouped in 4 categories as explained below which enable RAG models to be used in high-risk/high-interaction industries.

In a hybrid RAG arrchitecture, the LLM can operate at the natural-language input layer while OLM controls reasoning and output where determinism, traceability, and content control matter most. The operation is as simple as it is depicted in the diagram below.

In the first flow, the administrator (domain expert) is not required to have any experience in coding, prompting, ontological semantics, or AI. He/she sets up the RAG content to be handled by the system automatically.

In the second flow, the user can benefit from the system in a number of ways (called functions) ranging from question answering to knowledge mining. See the functions below.

OLM is the result of more than a decade of R&D in meaning representation, ontologies, deterministic AI, fuzzy logic, and machine reasoning. In 2026, the intellectual property of OLMapp was developed, enabling event-based ontologies to be acquired automatically from content and to be merged with a background ontology. This is the differentiator of our technology among all other ontological methods.

OLM was inspired by a series of historical developments. More than two millennia ago, Aristotle distinguished the symbols of individual languages from the underlying mental concepts and things they represent. OLM follows a modern computational objective: moving beyond linguistic symbols toward an explicit representation of meaning as outlined by the lineage below.

↓

↓

↓

↓

↓

LLMs cannot separate "language" from "knowledge". Because their training is exposed to knowledge while learning language (next token probabilities). For example, if the training material was heavily collected from Voo-doo-medicine, LLM's language would capture phrases of witchcraft advice. This is the basic reason of bias. LLMs predict plausible language; they do not inherently verify truth. That gap between Voo-doo medicine and modern medicine is where hallucinations arise.

OLM handles "language" and "knowledge" separately. One is not affected by the other. Event concepts represent knowledge of the world devoid of occurence statistics in any corpus. This results in the 4 categorical advantages listed at the top of the page.