You can use InfraNodus in a popular n8n automation tool to generate high-quality LLM responses based on an ontology that you specify in an InfraNodus graph.
The ontology can be created automatically or manually using the InfraNodus ontology generator. The resulting graph can then be queried using the official InfraNodus GraphRAG n8n node.
For instance, for this support portal's AI chat we use the following graph:
You can check this graph here:
https://infranodus.com/expert/infranodus_support
More reasoning ontology graphs are available at https://infranodus.com/knowledge-graphs
When a request from the user arrives to the graph via an n8n automation shown below, InfraNodus builds a graph from the user's query and finds overlapping structures in the ontology graph based both on graph search and vector similarity to retrieve the most relevant information (this is usually referred to as GraphRAG). The model will then ingest the underlying graph structure and produce a high-quality response.
Here is a link to this simple workflow on n8n: https://github.com/infranodus/n8n-infranodus-workflow-templates/blob/main/graphrag-chatbots-experts/graphrag-basic-ai-chatbot.json
In a more sophisticated setup, you can add an AI agent node that will orchestrate GraphRAG retrieval, combine it with search results from a knowledge base (e.g. your Zendesk support portal), maintain chat memory, and deliver the result to the user:
In this setup, you're using the InfraNodus expert ontology node to guide the model's thinking across a specific ontology graph (a part that relates to the user's query). It then augments the search query to Zendesk support portal to retrieve knowledge base articles. The results are then assembled both from Zendesk and reasoning ontology to provide the final response to the user.
Here's a link to this more sophisticated workflow template: https://github.com/infranodus/n8n-infranodus-workflow-templates/blob/main/customer-support-email/zendesk-ai-chatbot-agent.json
This same ontology graph can also be queried inside InfraNodus itself using the built-in InfraNodus AI widget or using the InfraNodus MCP server.
How is the Ontology Graph Created?
The graph contains relations between the entities / functions / features of InfraNodus (denoted with [[wikilinks]] and each relation has a description (denoted with [squarebrackets]). E.g.: "[[infranodus]] can be used for [[text analysis]] and finding [[content gaps]] to generate [[insights]] [usedFor]
The entities are the relations and if they appear in the same statement they are connected. We build a graph based on this structure and apply network analysis metrics to identify the main topical clusters and most influential nodes. You can learn more about the methodology in our peer-reviewed paper.
This structure is then converted into a text format and is used to augment LLM responses as the model is always provided the full context and also the underlying relations.
This approach, also known as GraphRAG, is much better than the traditional RAG (retrieval-augmented generation), because it takes the whole context and relations into account in addition to performing vector-based similarity search.
You can build the ontology manually or using the automated ontology creator in InfraNodus. We also have a Claude Skill that can generate an ontology for you that you can install and use in combination with the InfraNodus MCP server in Claude so that your LLM creates the ontology for you based on the data you provide and saves it to your graph. Learn more at Generate Knowledge Graphs and Ontologies in Plain Text.
Practical Use Cases for Reasoning Ontology
Here are some practical examples that demonstrate how the reasoning ontology is used to provide better responses to the users.
-
Nodus Labs Support Portal
We use the ontology graph https://infranodus.com/expert/infranodus_support to augment the n8n workflow that retrieves the results from our Zendesk support portal and provides the responses to the user. You can see how this chat widget works on the front page of https://support.noduslabs.com or using the popup button at the bottom right of this page.
Also, compare it to the native Zendesk AI widget available at https://infranodus.com to see which results you like better.
-
Circadian Books Website
In this example, a small independent publisher in Berlin — https://circadian.co — uses the InfraNodus graphs to help their readers speak to their books. Each book at the store has an AI chat which links directly to the relevant InfraNodus graph that contains information and the main relations from the book. These graphs are not based on entities but rather use very granular lemma-based representation, which actually works pretty well for conversational AI.
You can see it in action in this video:
Comments
0 comments
Please sign in to leave a comment.