Represent
We identify the concepts and relations of your domain. The agents propose the model; your experts validate it.
Shared concepts before content01 / VISIONAn AI that can tell you
An AI that can tell you
how it knows.
And that, when it matters, can say: I don't know
When an answer enters an organization's work, it must be open to examination: what information it uses, what relations it connects, what elements are missing. A trustworthy intelligence can show where an answer comes from, and knows to stay silent when the knowledge at its disposal is silent.
That is why we build an explicit representation of your domain: concepts and relations that your experts can discuss and correct. It is what we call an ontology — the description of what exists in your world, what it is called and how it connects to everything else.
From this idea follows a precise division of labor. Language models get the job they do best: understanding language, conversing, putting things into words. The content stays elsewhere, in a knowledge base that can be consulted and updated, on which answers are grounded and their steps verified.
It is an ancient idea. Porphyry's Isagoge, an introduction to Aristotle's Categories, is the first attempt to order knowledge so that it can be thought. Our name comes from there.
THE CASE OF A MUSEUMFrom the exhibition record
From the exhibition record
to the visitor's question.
Connecting exhibitions, works, authors and texts helps build an assistant that guides the public and makes the museum's wealth of information searchable.
02 / METHODAnalysis to the agents.
Analysis to the agents.
Judgment to the experts.
From your documents and your data, knowledge that reasons. In days, under your supervision.
Isagog has a method for building an organization's knowledge base from what it already has, and for reasoning over it. Our agents, based on specialized language models, read documents and data, recognize implicit concepts and relations and propose models of them, ready for review; other agents then extract structured information, item by item, in a traceable way. These are repeatable processes: new documents, new data, same procedures.
Domain experts keep the task only they can perform: supervision. They read what the agents have proposed, correct, approve — what used to take months of analysis and interviews is achieved in days. This is what makes the method scalable and keeps its costs under control.
Collect
Following the shared model, the agents extract information and link it to its sources. New documents feed the knowledge through repeatable procedures.
New data, a reusable methodReason
The language model brings the understanding of language; the knowledge graph brings structure, consistency and proof. It is the integration we call neurosymbolic, and it is the heart of our method.
Knowledge in everyday processesThis division of labor has an important consequence: when the content lives in the graph and the model takes care of language, even a small model is up to the task. It reasons on what is written, and what is written can be read, corrected and approved, without training or retraining anything.