GraphRAG: Building a Smarter AI System (full walkthrough)
📈 Get started with SerpApi for free 👉 https://serpapi.link/thu-vu💻 Git repo for this tutorial 👉 https://github.com/thu-vu92/graphRAG📝 Original Microsoft...
A Conceptual Data Model (CDM) is a high-level representation of business concepts and their relationships, independent of any technical implementation
This fourth edition of the newsletter is dedicated to conceptual data modeling for AI. A Conceptual Data Model (CDM) is a high-level representation of business concepts and their relationships, independent of any technical implementation.
0 notifications total
{"data":{"entityUrn":"urn:li:collectionResponse:4HlijJjSiEXmgxSNTaZbyKnZ88db1eTefgOnimBx6wM=","elements":[{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":2,"experimentId":5050470,"treatmentIndex":1,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"chameleonConfigTrackingItem":{"configLixTrackingInfoListV2":[{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":2,"experimentId":5050470,"treatmentIndex":1,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixTreatment":"VAR_t79530_PR_1","lixKey":"chameleon.SemanticJobSearch_Global.web-copy-definition.79529.child.79530","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfoWrapper"},{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":2,"experimentId":5050475,"treatmentIndex":1,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixTreatment":"VAR_t79529_PR_1","lixKey":"chameleon.SemanticJobSearch_Global.web-copy-definition.79529.child.79530","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfoWrapper"}],"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigTrackingItem"},"data":{"namespace":"jobs-semantic-search/components/educational-module","message":"I am looking for sales jobs in healthcare","locale":"en_US","key":"i18n_example_1","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigDataI18n"},"parentLixKey":"chameleon.SemanticJobSearch_Global.web-copy-definition.79529.parent","lixTreatment":"VAR_t79530_PR_1","parentLixTracking":{"urn":"urn:li:member:13500235","segmentIndex":2,"experimentId":5050475,"treatmentIndex":1,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixKey":"chameleon.SemanticJobSearch_Global.web-copy-definition.79529.child.79530","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigItem"},{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":3,"experimentId":5182155,"treatmentIndex":0,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"chameleonConfigTrackingItem":{"configLixTrackingInfoListV2":[{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":3,"experimentId":5182155,"treatmentIndex":0,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixTreatment":"control","lixKey":"chameleon.GLOBAL_premium.web-copy-definition.92114.child.92115","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfoWrapper"},{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":3,"experimentId":5182157,"treatmentIndex":0,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixTreatment":"control","lixKey":"chameleon.GLOBAL_premium.web-copy-definition.92114.child.92115","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfoWrapper"}],"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigTrackingItem"},"data":{"namespace":"organization-admin/components/edit-modal-tab-panels/premium-pages/manage-credibility-items","message":"","locale":"en_US","key":"i18n_description_v2","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigDataI18n"},"parentLixKey":"chameleon.GLOBAL_premium.web-copy-definition.92114.parent","lixTreatment":"control","parentLixTracking":{"urn":"urn:li:member:13500235","segmentIndex":3,"experimentId":5182157,"treatmentIndex":0,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixKey":"chameleon.GLOBAL_premium.web-copy-definition.92114.child.92115","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigItem"},{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":4,"experimentId":5001772,"treatmentIndex":2,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"chameleonConfigTrackingItem":{"configLixTrackingInfoListV2":[{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":4,"experimentId":5001772,"treatmentIndex":2,"$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfo"},"lixTreatment":"VAR_t73404_PR_2","lixKey":"chameleon.TGVerifications_GLOBAL.web-copy-definition.73402.child.73404","$type":"com.linkedin.voyager.dash.segments.chameleon.ChameleonConfigLixTrackingInfoWrapper"},{"lixTracking":{"urn":"urn:li:member:13500235","segmentIndex":4,"experimentId":5001773,"treatmentIndex":2,"$type":"com.linkedin.voyager.dash.segments.chameleon.Cham
LATRACE-AI: Multimodal memory service for text, audio, image, and video with tenant-isolated typed graph retrieval.
Multimodal memory service for text, audio, image, and video with tenant-isolated typed graph retrieval. - ZXXZ1000/LATRACE-AI
Unifying Ontology Construction and Semantic Alignment for...
While enterprises amass vast quantities of data, much of it remains chaotic and effectively dormant, preventing decision-making based on comprehensive information. Existing neuro-symbolic approaches rely on disjoint pipelines and struggle with error propagation. We introduce the large ontology model (LOM), a unified framework that seamlessly integrates ontology construction, semantic alignment, and logical reasoning into a single end-to-end architecture. LOM employs a construct-align-reason (CAR) pipeline, leveraging its unified architecture across all three stages: it first autonomously constructs a domain-specific ontological universe from raw data, then aligns neural generation with this structural reality using a graph-aware encoder and reinforcement learning, and finally executes deterministic reasoning over the constructed topology, node attributes and relation types. We evaluate LOM on a comprehensive benchmark constructed from diverse real-world enterprise datasets. Experimental results demonstrate that LOM-4B achieves 88.8% accuracy in ontology completion and 94% in complex graph reasoning tasks, significantly outperforming state-of-the-art LLMs. These findings validate that autonomous logical construction is essential for achieving deterministic, enterprise-grade intelligence.
Putting it all together: A demonstrator for knowledge engineering
Part 5 of my series on from Data Engineering to Knowledge Engineering
When Predicates Lie: SHACL, Reification, and the Event-First Ontology
Kurt Cagle and Chloe Shannon
ndoli: A personal knowledge graph (second brain) for Claude Code — tracks your professional network, projects, and opportunities using semantic web standards (RDF/OWL/SHACL).
A personal knowledge graph (second brain) for Claude Code — tracks your professional network, projects, and opportunities using semantic web standards (RDF/OWL/SHACL). - SteveHedden/ndoli
Beyond Graphify: Why the Enterprise Needs More Than a Folder-to-Graph Tool - AllegroGraph
There has been a lot of excitement lately around ideas like the “LLM wiki” and tools such as Graphify. The appeal is easy to understand. Instead of forcing an LLM … Continue reading Beyond Graphify: Why the Enterprise Needs More Than a Folder-to-Graph Tool →
Why Machines Need Embeddings: Turning Graph Structure into Features
This is Part 5 in a 5-part series on Graph Algorithms: Congratulations. We’ve done it. We have worked through four different algorithms, using intuitive visuals and easy-to-understand metaphors. But we have finally hit a roadblock. Now we must discuss embeddings,… Read more →
Understand-Anything: Turn any code, or knowledge base (Karpathy LLM wiki), into an interactive knowledge graph you can explore, search, and ask questions about. Graphs that teach graphs that impress. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
Turn any code, or knowledge base (Karpathy LLM wiki), into an interactive knowledge graph you can explore, search, and ask questions about. Graphs that teach > graphs that impress. Works with Cl...
Semantic Web Market Research Report 2025-2030: Semantic AI Convergence Unlocks Real-Time Analytics, Autonomous Systems and Intelligent Automation
The global semantic web market is set to expand from USD 2.71 billion in 2025 to USD 7.73 billion by 2030, at a CAGR of 23.3%. This growth is fueled by...
Supercharging LLM Wiki with Knowledge Graphs: Build a Self-Evolving Research System
Last week, Andrej Karpathy introduced a powerful concept called LLM Wiki — a framework designed to turn your notes, papers, and data into a structured, queryable knowledge base. It’s a big step for...
Rowboat: Open-source AI coworker that turns work into a knowledge graph and acts on it
Everyone Has an Ontology Now. Almost Nobody Has an Ontology. | LinkedIn
Microsoft just announced Fabric IQ. At its core: an "Ontology item.
RDF 1.2 Candidate Recommendation is a real milestone, but not the finish line.
RDF 1.2 Candidate Recommendation is a real milestone, but not the finish line. W3C advanced RDF 1.2 Concepts and Abstract Data Model and RDF 1.2 Semantics to Candidate Recommendation Snapshot on 7 April 2026, with comments requested by 5 May 2026.
For the Labelled Property Graph (LPG) world, the significance is not just timing. It is the continued narrowing of a long-standing modelling gap. RDF 1.2 standardises triple terms, reifying triples via rdf:reifies, annotation syntax, and clearer guidance on interoperability between RDF 1.2 Full and RDF 1.2 Basic. In practical terms, that gives RDF a stronger and more coherent way to attach metadata to statements such as provenance, confidence, validity interval, source system, policy context, or processing state.
That matters in enterprise settings because these are exactly the attributes that determine whether graph data can be trusted, governed, exchanged, and audited across organisational boundaries. In that sense, RDF 1.2 improves RDF’s role as a semantic contract and interchange layer.
But it still does not make RDF a native property graph. Triple terms denote propositions. They are typically surfaced through reifiers and semantic constructs, not as first-class operational edges with property bags in the LPG sense. That distinction remains important.
In most enterprises, LPGs earn their place because they are operationally direct. Teams use them to model application-facing networks, fraud paths, supply chains, identity graphs, recommendations, infrastructure dependencies, and decision flows where developers want explicit nodes, explicit relationships, straightforward mutation semantics, and predictable query patterns. Relationship properties are not an afterthought there. They are part of the working model.
This is also where the Applied Knowledge Graph (AKG) perspective matters. AKG is about using graph structures to support operational and analytical outcomes in the enterprise. That means combining semantics, governance, lineage, and domain meaning with execution models that are usable by engineering teams and close to business processes. In that architecture, RDF and LPGs are often complementary rather than mutually exclusive.
So the real impact of RDF 1.2 is not that LPGs become obsolete. It is that the boundary between semantic graph standards and operational graph platforms becomes easier to manage. RDF becomes better suited for rich meaning, controlled exchange, and durable interoperability. Labelled Property Graphs remain strong as the execution model for enterprise graph workloads that need speed, flexibility, and close alignment with business applications.
That is why I see RDF 1.2 as convergence, not replacement. Enterprises will still need LPGs for operational graph systems. What RDF 1.2 improves is the quality of the bridge: better semantics around statements, better interoperability options, and a stronger foundation for AKGs that need both enterprise control and practical delivery.
RDF 1.2 Candidate Recommendation is a real milestone, but not the finish line.
ETL & ELT
Column Operations & Graph Traversals
A good ontology is the outcome of an ontological analysis and meaning negotiation, making ontological commitments explicit
The term "ontology" has become overloaded. The meaning of ontology has turned into something like "whatever description encoded in Semantic Web languages" (RDFS, OWL, SHACL).
Why was this a mistake, not merely a terminology issue? It obfuscates the history of ontology in philosophy and its value as a toolbox for making **ontological commitments explicit**, rather than just a formal encoding of vague terminology. For example, you can create an OWL class Person to represent people involved in a mortgage contract. However, this encoding is just a mnemonic for many implicit assumptions. We're not clarifying anything here. We need more than formal semantics. We need real-world semantics, provided by foundational ontologies and ontological analyses. For example, a Person can be understood as (a) a human being; (b) a cognitively capable person; (c) a legally capable person; (d) a legal person; (e) a living human being, etc. Each of these concepts has a very different nature, and its instances behave differently.
The area of Applied Ontology (https://lnkd.in/ed8ZfJZP) is devoted to addressing these issues by developing theories based on which we can build domain ontologies that have explanatory power, in whatever language you decide to pick.
Ontology development is consensus creation, not (merely) representation (https://lnkd.in/evgfSpby). A good ontology is the outcome of an ontological analysis and meaning negotiation, making ontological commitments explicit. This cannot be extracted directly from documents because too many assumptions are implicit and reside in people's minds. Moreover, there isn't a unique way to do that. Ontologies are meaning contracts (https://lnkd.in/eyT9znTR).
The fundamental problem has never been a matter of representation language.
Y.digital, Semantics, Cybersecurity, and Services (SCS), University of Twente | 22 comments on LinkedIn
A good ontology is the outcome of an ontological analysis and meaning negotiation, making ontological commitments explicit
Integrate postgres and graphRAG in a docker, add AGE and AI agents. · GitHub
Integrate postgres and graphRAG in a docker, add AGE and AI agents. - Azure-Samples/postgreSQL-graphRAG-docker
OrionBelt Semantic Layer v1.3.0 is out, adding an ontology layer on top of the semantic layer
https://github.com/ralfbecher/orionbelt-semantic-layer/blob/main/ontology/spec.md https://github.com/ralfbecher/orionbelt-semantic-layer
OrionBelt Semantic Layer v1.3.0 is out, and in the Year of the Ontology, we're adding an ontology layer on top of the semantic layer
microsoft Ontology-Playground: Free, open-source web app for learning about ontologies and Microsoft Fabric IQ. Explore a catalogue of pre-built ontologies, design your own visually, export as RDF/XML, and share interactive diagrams. Zero backend, fully static.
Free, open-source web app for learning about ontologies and Microsoft Fabric IQ. Explore a catalogue of pre-built ontologies, design your own visually, export as RDF/XML, and share interactive diag...
RDF 1.2 Concepts, Abstract Data Model, Semantics
https://www.w3.org/TR/2026/CR-rdf12-semantics-20260407/ https://www.w3.org/TR/2026/CR-rdf12-concepts-20260407/
RDF 1.2 Concepts and Abstract Data Model
RVM: The Virtual Machine Reimagined for the Agentic Age. It models the system as a graph
RVM: The Virtual Machine Reimagined for the Agentic Age
Metadata Is the New Code: Structured Knowledge for AI-Driven
Most AI-assisted builds re-explain the same domain knowledge from scratch every session. Learn why reusable, versioned taxonomies and ontologies are t
LLM Wiki extended with agentmemory, a persistent memory engine for AI coding agents
Four failure modes appear in every enterprise knowledge graph. The missing layer is graph ontology
Hannah Stulberg built a Claude Code repo at DoorDash that 20 people use daily. Root CLAUDE.md under 500 tokens. Nested indexes in every folder. Each function owns its context. Data scientists pull metrics without asking. Strategy leads grab call summaries. Engineers check analytics without filing tickets.
She calls it a Team OS.
It is the most production-ready multi-user AI knowledge architecture I have seen documented publicly.
Here is the problem she will face within six months.
Andrej Karpathy 's LLM Wiki works beautifully for one person. You dump messy notes, the LLM compiles them into a wiki, every query makes the wiki better. The loop compounds because one mind, one vocabulary, one namespace. No ambiguity about what "revenue" means when you are the only person defining it.
Four failure modes appeared in every enterprise knowledge graph project since the Semantic Web era.
Ungoverned documents that make the graph a liability. Nobody owning the ontology so definitions drift silently. Semantic ambiguity where the same label means different things across functions. Maintenance decay where the system was created once and maintained never.
Stulberg's Team OS solves the first two with folder ownership: PM owns /product, data scientist owns /analytics, strategy owns /customer. Each function maintains its domain context. That is governance through file structure.
But folder ownership does not solve semantic ambiguity.
When the PM writes "conversion" in /product and the data scientist writes "conversion" in /analytics, those reference different metrics.
When engineering references a "service" and strategy references a "service," those are different entities.
The file system has no way to express that two identical labels point to different concepts. Flat markdown cannot represent typed relationships between entities.
The missing layer is a graph ontology.
Entities need types. Relationships need labels. Definitions need namespaces. A graph where "conversion" is a node with typed edges to its definition, its owner, its data source, and its measurement methodology prevents the silent drift that kills knowledge systems at team scale.
GraphQLite makes this practical. A single SQLite file with Cypher queries, schema-free node creation, and built-in algorithms like PageRank and Louvain community detection. No server. No configuration. Drop it into the Team OS repo alongside the CLAUDE.md files. When the agent needs to disambiguate "conversion," it queries the graph instead of guessing from context.
LightRAG takes this further for retrieval. Its dual-level system searches entities by vector similarity AND traverses their relationships in the graph. Legal document retrieval with graph-enhanced search scored 84.8% versus 15.2% for naive vector search. The graph does not replace the wiki. The graph makes the wiki queryable at the relationship level.
Four failure modes appeared in every enterprise knowledge graph
GEAKG: Generative Executable Algorithm Knowledge Graphs
In the context of algorithms for problem solving, procedural knowledge -- the know-how of algorithm design and operator composition -- remains implicit in code, lost between runs, and must be...
Lakehouse 2.0: Semantics as a first class citizen
seman
Common Beginner Mistakes in Ontology Projects (and how to avoid them)
Common Beginner Mistakes in Ontology Projects (and how to avoid them)
I often see enthusiastic starts in ontology engineering, followed by avoidable missteps that limit impact and reuse of ontology.
Here are a few patterns worth reflecting on:
1. Starting with tools instead of purpose
It’s tempting to jump straight into editors like Protégé or other modeling tools.
But ontology work isn’t tool-driven—it’s problem-driven.
Before choosing a tool, ask:
1.1 Where will this ontology be used?
1.2 Who are the users?
1.3 What decisions or queries should it support?
Tools don’t define good ontologies—clarity of purpose does.
________________________________________
2. Skipping competency questions
Competency questions are not just a formality—they are your design compass.
They define:
2.1 the scope of your ontology
2.2 how deep or broad it should go
2.3 which use cases matter
2.4 what problems you’re solving
Without them, you risk building something that:
a) isn’t reusable
b) doesn’t support real queries
c) becomes another silo in the semantic ecosystem
A simple rule:
If your ontology can’t answer its competency questions, it’s not done.
________________________________________
3. Over-modeling (the silent trap)
More detail ≠ more value.
Over-modeling often shows up as:
3.1 unnecessary class hierarchies
3.2 excessive axioms “just in case”
3.3 modeling edge cases before core use cases
This leads to harder maintenance, lower adoption and confusion for users and developers
Instead, aim for:
minimal viable ontology → model what is needed now, evolve later.
________________________________________
4. Ignoring domain experts
Ontology engineering is not a solo activity.
Domain experts bring:
4.1 real-world context
4.2 implicit knowledge
4.3 validation of concepts and relationships
As ontologists, our role is to:
a) translate domain knowledge into structured, interoperable representations
b) align with FAIR principles (Findable, Accessible, Interoperable, Reusable)
Without domain experts, even a “technically perfect” ontology can miss the mark.
Final thought
Good ontologies are not just well-structured—they are useful, usable, and used.
I’m curious to know what mistakes have you seen (or made) in early ontology projects?
If you're starting out in ontology engineering or transitioning into it, I’m happy to connect and exchange ideas.
Common Beginner Mistakes in Ontology Projects (and how to avoid them)
GraphQLite. An SQLite extension that adds graph database capabilities using the Cypher query language.
Ontology-driven cultural heritage on Databricks | LinkedIn
How maplib bridges tabular data, ontologies and open standards Riksantikvaren (Norwegian Directorate for Cultural Heritage) is pushing boundaries in how public institutions manage and share heritage data. Embracing semantic web technologies, open-source tooling, and international standards to future