Name tag
A document is cut into small, useful pieces called chunks. Each chunk keeps one unique ID and its source page.
Use case · Ontology-grounded enterprise AI
Give AI a map.
Not a pile of files.
An LLM is good at words. Your enterprise ontology teaches it what those words mean inside your organization—who owns what, how facts connect, which rules apply, what changed, and where the evidence came from.
One chunk identityVector + graph fusionEvidence on every answerAuthorization-aware
SHARED MEANINGEnterprise
OntologyAI context layer
OntologyAI context layer
01Datafacts + chunks
02Peopleroles + skills
03ServicesAPIs + lineage
04AIanswers + actions
The second-grade explanation
Every paragraph gets a name tag and a map pin.
Meaning magnet
A vector helps AI find chunks that sound like the question—even when they use different words.
Business map
The ontology and graph show what the chunk is about and how it connects to people, products, services, rules and risks.
Two clues become one
Vector clues and graph clues are ranked together. The LLM receives only the best authorized evidence.
Interactive enterprise GraphRAG story
Watch your data become
context an LLM can trust.
Choose a question, press Play, and follow the same evidence chunk through ingestion, vector search, the enterprise graph, fusion and the final answer.
ASK THE ORGANIZATIONWhy did the ETF signal change, and which portfolios are affected?
Bring the information in safely
We collect documents and data, but we never forget where each fact came from.
Batch, streaming and document-AI pipelines preserve source IDs, timestamps, access scope and the original payload.
THIS STEP CREATESSource records with time, owner and permissions
The architectural breakthrough
One chunk. Two retrieval superpowers.
The same stable chunk ID exists in both stores, so the answer never loses its connection to source evidence.
“The issuer lowered full-year guidance after reporting weaker demand.”
Document: 8-KPage: 3Bounding box preserved
Find similar meaning
The embedding helps retrieve wording that resembles the question.
CH-104 → [0.18, 0.72, …]Follow business relationships
The evidence node links the chunk to the issuer, guidance event, ETF, portfolio and risk rule.
CH-104 → Event → ETF → PortfolioHow AI uses ontology at different levels
A reusable agent platform.
Domain meaning changes—not the plumbing.
The enterprise ontology supplies a runtime profile so shared execution agents can work safely across finance, cyber, skills, operations and other domains.
Orchestrator
Receives the question and sends it to the correct domain flow.
“Who should answer this?”Domain coordinator
Uses ontology to classify intent, scope concepts, choose relationships, apply policy and build the retrieval plan.
“What does this question mean here?”Execution agents
Reusable vector, graph and Text2SPARQL executors fetch candidate chunks, facts and paths.
“Bring back the right clues.”Evidence experience
Renders the answer and turns trusted evidence metadata into clickable source pages and highlights.
“Show me why I should trust it.”Shared meaning across the organization
People and platforms stop
translating the same thing again.
Coworkers & skills
Connect people, validated skills, certifications, project experience, availability and role requirements—with privacy and authorization controls.
Business services
Trace a service to the APIs, data products, applications, owners, controls and business outcomes it depends on.
Data pipelines
Share the same definitions and identities from ingestion through transformation, quality, lineage and downstream consumption.
AI agents
Receive authorized facts, rules, evidence and uncertainty instead of searching an ungoverned pile of text.
Risk & compliance
Connect obligations and controls to the exact processes, models, assets, decisions and evidence they govern.
Technology platforms
Expose shared meaning through semantic APIs so catalogs, data products, graph tools, search and applications agree.
How Semantic Data Services gets you there
From one important question
to an AI-ready semantic platform.
Senior ontology engineers work with business, data, security and AI teams, then transfer the capability so your organization can own it.
DISCOVER
Choose the business questions
Identify decisions, evidence, users, risks and measurable outcomes.
MODEL
Build enterprise and domain ontologies
Define shared concepts, relationships, rules, identities, time and provenance.
CONNECT
Map sources and chunks
Align structured fields and unstructured evidence to governed meaning.
POPULATE
Create graph and vector twins
Use stable identity across embeddings, RDF facts, evidence nodes and source metadata.
ACTIVATE
Build GraphRAG and agent context services
Implement routing, SPARQL, vector retrieval, centrality ranking, fusion and grounded prompts.
GOVERN
Secure, evaluate and operate
Apply access policy, SHACL, golden questions, citation checks, observability and OntologyOps.
Ontology is a boon—not magic
The LLM writes the sentence.
Your enterprise knowledge supplies the truth boundary.
Ontology improves understanding, retrieval and consistency. Production AI still needs approved data, access controls, evaluation, evidence-quality checks, human review and a deterministic “not enough evidence” response.
✓ No invented evidence links✓ Same chunk identity across stores✓ Claim-to-citation consistency✓ Golden-question evaluation✓ Permission-aware context✓ Human-review boundary
Technical foundations
This implementation pattern is grounded in RDF/SPARQL graph retrieval, provenance modeling, hybrid vector search and reciprocal-rank fusion.
Start with one question
Let your AI understand
how your organization works.
We will trace the required data, chunks, concepts, relationships, rules, permissions and evidence into a practical GraphRAG pilot.