GraphNews

Vendors want to sell you context graphs. You deserve a way to evaluate that claim. Knowledge graphs: Define what you know Context graphs: Define as you go Right now procurement teams can check SOC 2, uptime, security certs. You cannot yet check whether a vendor knows the difference between a verified system boundary and one that was never measured. One gap the W3C Context Graph Community Group exists to close. This week some vendors got upset with me for drawing a line. Let me explain why the line matters — and why it matters to you, not just to us. Semantics needs authority: controls, ontology, governance of vocabulary. That's its power - and its limit. If you are a vendor, or the employee owning it: this is your incentive structure. The moment something happens outside that control — same word, different business unit, acquired company, connected system — there's a conflict. Knowledge graphs don't cover that space & they can't. People are building knowledge graphs about context and calling them context graphs. An LLM that retrieves from one and calls it "context" is still RAG. Rebranding doesn't change the instrument. A context graph is for everything outside the boundary. Semantics: you pay to control things. Context: you don't — or can't — control things. Non-technical people with AI are creating code with data, touching systems, knowing nothing about ontologies. You can't impose a top-down vocabulary on how everyone thinks. Something has to navigate the space between systems — coherently, measurably. That's what a context graph does. Not competing with knowledge graphs. Not even the same territory. A knowledge graph is designed toward completeness. A context graph maps the space outside any one system's boundary. A knowledge graph requires shared governance. A context graph requires a shared handshake. Both needed. Not competing. After I point this out, semantic vendors accuse me of co-opting the word context. In our O'Reilly book Unifying Business, Data, and Code — which Juan and I started three years ago — context was one of four foundational pillars, and the last chapter applied it to execution-time graphs. I find it ironic that a community built on provenance didn't check mine before telling me what context means. Shannon had to co-opt "information" and give it a precise mathematical definition. We're doing the same for "context" — because in banking we found severe vulnerabilities no existing framework covers. They live at system boundaries, where the reference frame stays behind when data crosses. Precision isn't a turf grab. It's what makes evaluation possible. My bet on the future of semantics: Nicolas Figay. His work on projections & cartography is sidereal navigation for knowledge systems - directly aligned to our view. 61 participants: banking, healthcare, supply chain, and academia. Innovation program starts Tues! If you're evaluating vendors making context graph claims — I'm here to help.

If you are interested in an RDF Knowledge Graph support library, I have something for you :) rdf4cpp is a modern C++20 library providing basic RDF support. The focus is correctness, performance and ease-of-use for basic building blocks like: - parsing, validating and writing RDF data (N-Triples, Turtle, N-Quads, TriG) - parsing and validating RDF/XML - complete and extensible literal datatypes (validation, functions, operations, subtype and promotion casting, mapping to C++ types, error handling, ...) - managing RDF nodes efficiently - blank node scoping (e.g., for RDF datasets) rdf4cpp is not a SPARQL engine or reasoning engine, although it provides very basic support for triple/quad pattern matching on RDF graphs/datasets. rdf4cpp rather provides the necessary primitives to implement such engines. -- check it out: https://lnkd.in/eXCcz35R and if you like it, don't forget to give a star 🥳

In this video, I break down the key property characteristics that control how relationships behave in an ontology or knowledge graph. From symmetric, asymmet...

What actually is a context graph? James Kaplan (McKinsey Technology CTO): a knowledge graph with decision traces: state, plus the actions that led to state. Animesh Koratana (PlayerZero CEO): knowledge graph is over-scoped. It doesn't even have to be a graph. What it has to do is model relationships between data across systems, scoped to a specific problem, because the relationship between two records changes depending on the decision you're making. Full discussion → https://lnkd.in/g6enxRsS

Authored by Animesh Koratana, founder and CEO of PlayerZero We recently wrote about context graphs, the layer that captures decision traces rather than just data. The argument: the next trillion-dollar platforms won't be built by adding AI to existing systems of record, but by capturing the reasonin

LLMs are powerful, but without a semantic backbone, they are just sophisticated guessers. Here’s why the future of enterprise AI is built on formally structured knowledge.

Legal texts form a dense network of cross-references, definitions, and obligations. This makes it hard to find and understand how a specific requirement applies to a company's activities. That is why for the last few months I've been working on a legal tech project to turn the law into a knowledge graph with agentic capabilities. The knowledge graph consists of three subgraphs: - The law itself with a hierarchical structure - The company context and activities  - A manually created analysis layer that maps how specific legal obligations apply to a company's activities given its context We can then search over this graph to answer legal questions where each statement traces back to a legal source and its human interpretation in the context. This also allows you to see the connections between different requirements and explore deeper as needed, with or without AI. If you deal with legal or compliance questions in your work, I'd love to talk with you and understand the kind of challenges you face. I'd be happy to share my insights or walk you through a demo.

Introducing: PlayerZero. The world's first Engineering World Model that puts debugging, fixing, and testing your code on autopilot. We've raised $20M from Foundation Capital, and the founders of Figma, Databricks, Vercel, and Dropbox. It frees up 30% of your engineering bandwidth by: 1.⁠ ⁠Finding the root cause for bugs & incidents in minutes that engineering teams take days to identify. 2.⁠ ⁠Predicting in minutes, edge case issues that a 300-person QA team would take weeks to find. ------ Here's why this matters: No one in your org has a complete picture of how your production software actually behaves. Support sees tickets. SRE sees infra. Dev sees code. Each team builds their own fragmented view - and none of these systems talk to each other. When something breaks, everyone scrambles to stitch the picture together by hand. PlayerZero connects all of it into a single context graph - → The Slack thread where your lead said "we went with X because Y fell apart in prod last time" → The PR review where an engineer explained the tradeoff → The lifetime history of your CI/CD pipeline, observability stack, incidents, and support tickets So you can trace any problem to its root cause across every silo. And it compounds. Every incident diagnosed teaches the model something new. The longer it runs, the deeper it understands - which code paths are high-risk, which configurations are fragile, which changes tend to break which customer flows. So when you sit down to debug a live issue, you have your entire org's collective reasoning and production memory behind you - instantly. ------ Zuora, Georgia-Pacific, and Nylas have reduced resolution time by 90% and caught 95% of breaking changes and freeing an average of $30M in engineering bandwidth. PlayerZero is the SOTA AI production engineer. Across 26,400 real PRs, PlayerZero predicted bugs before code shipped to production 64% of the time. For comparison: → Claude Code: 11% → Cursor BugBot: 16% ------ Our guarantee: If we can't increase your engineering bandwidth by at least 20% within one week, we'll donate $10,000 to an open-source project of your choice. Book a demo - https://bit.ly/3NlLMeN We're also giving away a curated collection of 200+ Claude Code skills our team uses daily - the workflows that made us faster engineers while building PlayerZero. Comment "100x" to get access. | 374 comments on LinkedIn

Where We Were, Where We Are, and the Future

Crack open your codebase RAG with a knowledge graph analysis that illustrates hidden dependencies and other structural breakdowns in part 2 of this series.

Making sense of the buzzwords most data teams are still trying to figure out

github.com/averdeny/graphglot

AI Shortcuts the Ontology Tax Saturday reflections. 🤔 Ontologies and knowledge graphs are so hot right now and I'm always keen to read good posts about the realities of implementing them. As such, I was intrigued to read Alexander Shereshevsky’s piece -“The Ontology Tax: What Nobody Tells You About the Real Cost of Knowledge Graphs” - (link in comments). It's operner is about the tax most don’t see coming. His headline: Typical cost, between $500K-$1.5M+ over five years for a single domain ontology. Months of expert work before usage with 15-25% annual maintenance. With governance adding 30-50% He says only 27% of orgs get knowledge graphs to production despite a $6.9B market. The scale is staggering. The SNOMED CT initiative- has approx. 350,000+ clinical concepts and requires an entire non-profit organisation just to maintain it with monthly releases. Semantics are super important for accurate and robust information based decisioning systems, especially if one is using AI. But with traditional manual approaches, typically there is this massive effort needed before any use-cases can be even started and also to get up-to-date. It’s interesting in my work building these types of systems I have found the same thing. Spending weeks trying to classify domain knowledge, build taxonomies and align definitions. Every time I added to the graph it became more brittle. A compliance process step isn’t an obligation or prohibition - the ontology had no box for it, so I spent ages trying to shoehorn into compromise approaches. This is what led me to creating a new type of small language model (now Nebulyx Cognition) that discovers semantic structure directly from language. It learns that “shall report all transactions” and “clinicians should offer CBT” share the same shape - someone must do something w/ a degree of authority. A really cool part is that business users can adjust, add and maintain through simple user interfaces. I.e. we don't require armies of knowledge specialists. But the grunt work is done by the model i.e. encode 70-80% automatically and directly from source material. When one queries e.g. “what are the medical procedures for x” - the actual text and location form the medical book is returned, every time. We can constantly and quickly update as new information comes in. A change in procedure or policy a new regulation. Plus, link and do differences between text by querying the AI - which is very tough to do with traditional methods. All within minutes. Knowledge compounds through this agile flywheel effect rather than drifting despite it. Semantic graphs (including Ontologies and Knowledge Graphs) are awesome and needed in an AI world, but we shouldn’t have to pay a large tax to adopt them Happy Saturday. Nebulyx AI | 32 comments on LinkedIn

RDF file extension for DuckDB. Reads and writes supported - nonodename/duck_rdf

I came across a solution that changes how Claude Code handles large codebases.

Large Language Models (LLMs) demonstrate strong reasoning abilities but face limitations such as hallucinations and outdated knowledge. Knowledge Graph (KG)-based Retrieval-Augmented Generation...

Why Hypergraphs? Most graph databases and modeling tools are built on simple graphs — nodes connected by edges, where each edge links exactly two nodes.

Mariam Barry, Gaetan Caillaut, Pierre Halftermeyer, Raheel Qader, Mehdi Mouayad, Fabrice Le Deit, Dimitri Cariolaro, Joseph Gesnouin. Proceedings of the Workshop on Generative AI and Knowledge Graphs (GenAIK). 2025.