Hellixia User Guide

Hellixia is BayesiaLab’s subject matter assistant, powered by Generative AI. It helps you draft, enrich, and review the parts of a model, from nodes and arcs to probabilities, comments, labels, and images, starting from a prompt, a set of documents, or a network you already have. The functions below are grouped by task and follow the order of the Hellixia menu.

Setup

Settings

Settings is where you connect BayesiaLab to a Generative AI provider so that every other Hellixia function can call a language or embedding model.

Enter your provider account, API key, model choice, and any custom endpoint, then verify the connection. You might choose an embedding model for semantic work and a separate completion model for text generation. Configure this once before running anything else; if a function reports a missing key or a provider error, this is the first place to check.

Generating Networks from a Prompt

These functions turn a written question or prompt into a first draft of a network, giving you a structured starting point that you then refine by hand.

Automatic Bayesian Network Generator

The Automatic Bayesian Network Generator turns a written description of a problem into a draft Bayesian network, proposing the nodes, arcs, and probability tables from your prompt.

Automatic Causal Network Generator

The Automatic Propositional Causal Bayesian Network Generator turns a written description of a problem into a draft Bayesian network: nodes joined by causal arcs and quantified with probability tables.

Give it a question or a short description of your domain, plus any background context or modeling constraints, and it proposes the nodes, the causal arcs with explanatory comments, causal effects, and conditional probability tables. From a prompt about car accident risk, for example, it can draft a network linking driver behavior, vehicle condition, road infrastructure, and weather. Treat the result as a first hypothesis and review the structure and the numbers against expert knowledge before using the model for analysis.

Automatic Semantic Network Generator

The Automatic Semantic Network Generator turns a question into a map of related concepts, inventing the relevant dimensions, generating data for them, and learning a network that places similar concepts close together.

Supply a question and optional keywords; it extracts semantic dimensions, computes embeddings (numerical representations of meaning), synthesizes a dataset, and learns a Semantic Network from it. Starting from a question about how consumers perceive cars, for example, it can surface themes such as safety, performance, and comfort and place related concepts near each other. Use it to explore a domain or surface themes early on, then validate the discovered structure before relying on it.

Conversational Network Modeler

The Conversational Network Modeler lets you build and refine a network through a guided conversation with the model, turning the exchange into nodes, arcs, and structure.

Document Analysis

Document Analysis is the entry point for turning uploaded documents, called Knowledge Files, into model structure, routing your source material to one of six specialized generators. Add one or more Knowledge Files (reports, articles, transcripts, or other text) and choose the generator that matches the artifact you want. Several of these generators are counterparts of functions listed elsewhere in this guide, except that each works from file content rather than from selected nodes or typed text.

Semantic Flowchart Generator

The Semantic Flowchart Generator reads a Knowledge File and lays out its key concepts as a flowchart, connecting them in the order the document describes.

It is best for narratives, procedures, and process descriptions where sequence matters, and it produces a Semantic Flowchart you can review against the source. Pointed at a philosophical text, for instance, it can trace a progression such as Sensory Certainty to Perception to Understanding to Reason.

Causal Semantic Diagram Generator

The Causal Semantic Diagram Generator reads a Knowledge File and joins its key concepts with causal arcs, capturing the causal mechanisms the text states or implies.

Use it to extract causal stories from documents before formal Bayesian modeling; it produces a Causal Semantic Diagram with explanatory arc comments for expert review. From a medical article on skin hyperpigmentation, for example, it can map genetic predisposition and environmental triggers through melanin dysregulation to visible darkening.

Knowledge Graph Generator

The Knowledge Graph Generator extracts the entities named in a Knowledge File and links them with labeled semantic relationships, producing a graph of entities and their connections.

Use it to pull entities and relationships out of a document or to explore how its concepts connect. Run across the sections of a long text such as Marx’s Paris Manuscripts, it can build one interconnected graph of the economic and philosophical concepts it raises.

Causal Network Generator

The Propositional Causal Bayesian Network Generator builds a quantified causal model, a Bayesian network or Risk-Centric Causal Network, directly from the content of a Knowledge File.

It proposes nodes, causal arcs with mechanism comments, causal effects, conditional probability tables, and prior probabilities for root causes. Given a technical or clinical report, for example, it can draft a working causal model that you then refine. Review the generated structure and numbers before operational use.

Semantic Network Generator

The Semantic Network Generator learns a Semantic Network from a Knowledge File by extracting latent dimensions and concept embeddings, revealing how the document’s ideas cluster by meaning.

Provide the file and selected keywords; it returns the extracted dimensions, embeddings, and a learned Semantic Network for thematic exploration of the source material. It is a good way to see which themes in a long report sit close to one another.

Doc-to-Node Generator

The Doc-to-Node Generator creates one node per Knowledge File and stores each document’s content in that node, so an entire collection becomes a network you can compare and map.

Select several Knowledge Files; it generates one node per document with the file content kept as a comment, plus optional embeddings and an optional Semantic Network linking the documents. Given a folder of articles or transcripts, for example, it can place similar documents near each other for comparison across the whole corpus.

Semantic and Relationship Analysis

These functions identify concepts, entities, and the relationships among them, working from selected nodes, highlighted text, or a typed prompt rather than from uploaded documents.

Semantic Flowchart Generator

The Semantic Flowchart Generator arranges concepts into a flowchart that shows their sequential order, working from selected nodes, their comments, or text you supply.

Use it for process descriptions, workflow narratives, and ordered explanations, such as the stages of a customer journey; it produces a Semantic Flowchart. For the version that reads uploaded documents, see Document Analysis > Semantic Flowchart Generator.

Causal Semantic Diagram Generator

The Causal Semantic Diagram Generator connects concepts with causal arcs to sketch a causal picture, working from selected nodes, their comments, or text you supply.

Use it to sketch causal mechanisms and generate hypotheses before formal modeling; it produces a Causal Semantic Diagram with explanatory arc comments for review. You might use it to lay out the factors you believe drive an outcome such as project delays.

Knowledge Graph Generator

The Knowledge Graph Generator builds a graph of concepts or entities joined by labeled semantic relationships, working from a prompt, selected text, or selected nodes.

Use it for relationship mapping, domain inventories, and conceptual exploration; it produces a Knowledge Graph. From a prompt, for example, it can map the major figures and ideas in a field such as Western philosophy.

Entity Relationship Finder

The Entity Relationship Finder examines a set of selected entities or nodes and proposes the semantic relationships that connect them, each with a short explanation.

Use it to explain why concepts are linked, to build candidate relationships, or to check specific pairs; it returns proposed relationships, their labels, and reviewable explanations. Given a set of philosophers as nodes, for example, it can propose who influenced whom.

Verbalize Relationships

Verbalize Relationships turns the arcs in your model into plain sentences that describe what each relationship means.

Select arcs or relationships and it writes natural language descriptions suitable for documentation, communication, or review. It can state, for example, that fueling delays lengthen aircraft turnaround time, in words a reader without the model can follow.

Causal Knowledge Mining and Priors

These functions elicit causal directions, mechanisms, effects, and probabilities for a set of nodes you already have, giving you reviewable causal hypotheses and starting numbers.

Causal Network Generator

The Propositional Causal Bayesian Network Generator assembles a full causal model, a Bayesian network with causal arcs and probability tables, from a prompt, selected nodes, or background context.

It outputs nodes, causal arcs, causal effects, conditional probability tables, and prior probabilities for root causes. From a prompt about flight delays, for example, it can draft a model linking weather, fueling delays, aircraft turnaround time, and departure delays. Use it to bootstrap a causal model or a Risk-Centric Causal Network, then edit before operational use. For the version that works from documents, see Document Analysis > Propositional Causal Bayesian Network Generator.

Causal Relationship Finder

The Causal Relationship Finder reviews a set of selected nodes and proposes which ones plausibly cause which, drawing each as a directed arc with an explanation.

Use it to screen for causal links, elicit mechanisms, and map dependencies among existing nodes; it returns proposed causal arcs, mechanism descriptions, and arc comments. Given the factors in a flight delay model, for example, it can propose which ones drive the others.

Multi-Engine Causal Relationship Finder

The Multi-Engine Causal Relationship Finder asks several different language models the same causal questions and compares their answers, so your conclusions do not rest on a single model.

Provide selected nodes and configure multiple LLM engines; it returns each engine’s proposals side by side as comparison material for expert review. You might run the same question about what drives overall liking of a car across Claude, GPT, Gemini, Mistral, and Llama, then check where they agree.

Causal Structural Priors

Causal Structural Priors proposes expert constraints on a network’s structure, namely which causal directions are plausible, that you can feed into BayesiaLab’s structure learning.

Select arcs or nodes and it returns structural priors, explanations of the suggested causal directions, and review material. It can turn an observed association, such as the one between New Deal policies and economic recovery, into a directional hypothesis you use to constrain learning. Two worked workflows are documented under this function.

Pairwise Causal Link

Pairwise Causal Link looks at exactly two nodes and proposes which way the causal arc should point between them, with an explanation.

Use it for focused checks between two nodes and for review at the arc level; it returns a proposed causal direction, a mechanism explanation, and an optional arc comment. You might ask whether fueling delays directly cause longer aircraft turnaround time.

Root Priors Elicitor

The Root Priors Elicitor estimates starting probabilities for root nodes, the nodes that have no causes of their own, so they hold sensible baseline values before any evidence is entered.

Provide the root nodes and their state definitions; it returns prior probability distributions and elicitation comments for expert review before inference. It can suggest, for example, the baseline probability that a car has prominent advanced safety systems.

ICI Local Effects Elicitor

The ICI Local Effects Elicitor estimates how strongly each cause independently pushes its effect, the parameters needed for Noisy-OR and ICI nodes, a compact way to define probabilities when several causes act independently.

Select the ICI relationships and node states; it returns independent local effect estimates and reviewable parameter notes, sparing you from filling in large probability tables by hand. It can estimate, for example, how much safety systems, interior luxury, and performance each raise overall liking of a car.

Authoring, Enrichment, Translation, and Visualization

These functions polish and document a model you already have, generating comments, labels, classes, embeddings, images, and translations.

Dimension Elicitor

The Dimension Elicitor suggests the relevant aspects, or dimensions, of a topic and creates a node for each, so you can scaffold a model from a few keywords.

Provide keywords and optionally selected nodes; it creates dimension nodes with descriptive comments and optional classes. From the topic of philosophy, for example, it can produce a node per major philosopher along with fields such as birth year and area of work. It is useful for survey item ideation and early scaffolding.

Textual Dimension Elicitor

The Textual Dimension Elicitor reads text you provide and proposes the relevant dimensions of the topic, creating a node for each.

Textual Imputation

Textual Imputation uses a language model to fill in missing values, inferring each from the surrounding textual context.

Embedding Generator

The Embedding Generator converts the text attached to nodes, their names, long names, or comments, into embeddings: numerical vectors that capture meaning so concepts can be compared and clustered.

Select nodes and choose which text to use; it produces embeddings you can use for learning semantic proximity, clustering, and building a Semantic Network. A node named Risk Tolerance with a short comment, for example, becomes a vector that can sit near related risk concepts.

Comments

Comments groups seven tools that write, expand, shorten, or derive the descriptive text attached to nodes and arcs. Use them to document a model, prepare explanations for readers, or clean up existing text; in a flight delay network, for example, they can describe dozens of ancestor and descendant nodes at once. Each tool below targets one specific comment task.

Keyword-Based Node Comment Generator

The Keyword-Based Node Comment Generator writes a node’s comment from keywords you supply, grounding the description in your chosen terms.

You might document a node named Risk Tolerance from keywords such as uncertainty and willingness to accept loss. It is good for fast documentation aligned to your vocabulary.

Definition-Based Comment Generator

The Definition-Based Comment Generator writes a node’s comment from a concept definition, producing precise documentation grounded in domain vocabulary.

Use it when you want every node described in the exact terms your field uses, which helps standardize terminology across a model.

Node Comment Condenser

The Node Comment Condenser shortens long node comments into concise summaries that fit a tooltip while keeping the essential meaning.

Point it at nodes whose comments have grown too long to read at a glance.

Node Comment Elaborator

The Node Comment Elaborator expands a brief node comment with added explanation, context, and examples.

Use it to turn a brief note into something a reader can follow without prior knowledge of the model.

Long Name Generator from Node Comment

The Long Name Generator from Node Comment derives a readable display label for a node from its comment, improving how the model reads on screen.

A node with a terse name but a rich comment, for example, can gain a clear long name that summarizes it at a glance.

Arc Comment Condenser

The Arc Comment Condenser shortens the comment on an arc while preserving the mechanism it describes.

Use it to tidy wordy notes on the links between nodes.

Arc Comment Elaborator

The Arc Comment Elaborator expands an arc’s comment with more detail about the causal or semantic mechanism linking two nodes.

A terse note on the arc from fueling delays to aircraft turnaround time, for example, can become a full explanation of how one delays the other.

Class Description Generator

The Class Description Generator writes a description for each class, a named group of nodes, and suggests an interpretive name that captures the group’s theme.

Select classes or grouped nodes; it returns class descriptions and suggested names. Given clusters of car features, for example, it can name latent factors such as Safety, Performance, Interior Design, and Comfort. It is useful for interpreting clusters and naming factors.

Semantic Variable Clustering

Semantic Variable Clustering groups nodes that mean similar things into clusters, using their names, long names, comments, or embeddings to organize a large network.

Select the nodes to organize; it returns semantic clusters and classes you can review. In a flight delay model, for example, it can gather many upstream nodes into clusters such as Weather Conditions and Crew Readiness. It is useful for tidying large networks or grouping survey items.

Image Generator

The Image Generator creates an illustrative picture for each selected node from its name or description and attaches the image to the node.

Select nodes and optionally guide the prompt; it generates images that aid recognition in presentations and communication. In a flight delay network, for example, it can give the weather node a cloud icon and the turnaround node a plane icon.

Translator

The Translator renders a model’s text, its node names, state names, and comments, into another language using a translation engine such as ChatGPT or DeepL.

Select the nodes and choose source and target languages; it returns translated labels, state names, and comments for multilingual models and documentation. You can, for example, convert an entire network from French to English while keeping its structure intact.

Workflow Guides

I Have a Question or Prompt

Start with Automatic Propositional Causal Bayesian Network Generator or Automatic Semantic Network Generator, then review the proposed structure.

I Have Documents

Use Document Analysis to turn Knowledge Files into graph artifacts. Common next steps include Document Analysis > Propositional Causal Bayesian Network Generator and Document Analysis > Doc-to-Node Generator.

I Already Have Nodes

Enrich selected nodes with Dimension Elicitor, Embedding Generator, Semantic Variable Clustering, or Comments.

I Need Causal Orientation or Priors

Use Causal Relationship Finder, Causal Structural Priors, Root Priors Elicitor, or ICI Local Effects Elicitor to create reviewable hypotheses.

I Need to Communicate or Enrich a Model

Use Verbalize Relationships, Class Description Generator, Image Generator, or Translator.

Troubleshooting

Examples and Tutorials

• Hellixia Examples

  Browse examples by domain, input type, output type, and workflow.

  →
• Webinar: Discovering Complex Causal Structures

  See Hellixia causal analysis functions in a webinar workflow.

  →
• Webinar: Harnessing Hellixia Innovations

  Review Bayesian network construction workflows with Hellixia.

  →
• Webinar: From Narratives to Networks

  Follow a workflow that turns narrative text into a causal network.

  →

Key Concepts

Before You Begin

Confirm the following before running a Hellixia function: