Knowledge Graphs and LLMs in Action - Alessandro Negro with Vlastimil Kus, Giuseppe Futia and Fabio Montagna
Knowledge graphs help understand relationships between the objects, events, situations, and concepts in your data so you can readily identify important patterns and make better decisions. This book provides tools and techniques for efficiently labeling data, modeling a knowledge graph, and using it to derive useful insights.
In Knowledge Graphs and LLMs in Action you will learn how to:
Model knowledge graphs with an iterative top-down approach based in business needs
Create a knowledge graph starting from ontologies, taxonomies, and structured data
Use machine learning algorithms to hone and complete your graphs
Build knowledge graphs from unstructured text data sources
Reason on the knowledge graph and apply machine learning algorithms
Move beyond analyzing data and start making decisions based on useful, contextual knowledge. The cutting-edge knowledge graphs (KG) approach puts that power in your hands. In Knowledge Graphs and LLMs in Action, you’ll discover the theory of knowledge graphs and learn how to build services that can demonstrate intelligent behavior. You’ll learn to create KGs from first principles and go hands-on to develop advisor applications for real-world domains like healthcare and finance.
Most people talk about AI agents like they’re already reliable. They aren’t.
Most people talk about AI agents like they’re already reliable. They aren’t.
They follow instructions. They spit out results. But they forget what they did, why it mattered, or how circumstances have changed. There’s no continuity. No memory. No grasp of unfolding context. Today’s agents can respond - but they can’t reflect, reason, or adapt over time.
OpenAI’s new cookbook Temporal Agents with Knowledge Graphs lays out just how limiting that is and offers a credible path forward. It introduces a new class of temporal agents: systems built not around isolated prompts, but around structured, persistent memory.
At the core is a knowledge graph that acts as an evolving world model - not a passive record, but a map of what happened, why it mattered, and what it connects to. This lets agents handle questions like:
“What changed since last week?”
“Why was this decision made?”
“What’s still pending and what’s blocking it?”
It’s an architectural shift that turns time, intent, and interdependence into first-class elements.
This mirrors Tony Seale’s argument about enterprise data: most data products don’t fail because of missing pipelines - they fail because they don’t align with how the business actually thinks. Data lives in tables and schemas. Business lives in concepts like churn, margin erosion, customer health, or risk exposure.
Tony’s answer is a business ontology: a formal, machine-readable layer that defines the language of the business and anchors data products to it. It’s a shift from structure to semantics - from warehouse to shared understanding.
That’s the same shift OpenAI is proposing for agents.
In both cases, what’s missing isn’t infrastructure. It’s interpretation.
The challenge isn’t access. It’s alignment.
If we want agents that behave reliably in real-world settings, it’s not enough to fine-tune them on PDFs or dump Slack threads into context windows. They need to be wired into shared ontologies - concept-level scaffolding like:
Who are our customers?
What defines success?
What risks are emerging, and how are they evolving?
The temporal knowledge graph becomes more than just memory. It becomes an interface - a structured bridge between reasoning and meaning.
This goes far beyond another agent orchestration blueprint. It points to something deeper: Without time and meaning, there is no true delegation.
We don’t need agents that mimic tasks.
We need agents that internalise context and navigate change.
That means building systems that don’t just handle data, but understand how it fits into the changing world we care about.
OpenAI’s temporal memory graphs and Tony’s business ontologies aren’t separate ideas. They’re converging on the same missing layer:
AI that reasons in the language of time and meaning.
H/T Vin Vashishta for the pointer to the OpenAI cookbook, and image nicked from Tony (as usual). | 72 comments on LinkedIn
Most people talk about AI agents like they’re already reliable. They aren’t.
A gentle introduction to DSPy for graph data enrichment | Kuzu
📢 Check out our latest blog post by Prashanth Rao, where we introduce the DSPy framework to help you build composable pipelines with LLMs and graphs. In the post, we dive into a fascinating dataset of Nobel laureates and their mentorship networks for a data enrichment task. 👇🏽
✅ The source data that contains the tree structures is enriched with data from the official Nobel Prize API.
✅ We showcase a 2-step methodology that combines the benefits of Kuzu's vector search capabilities with DSPy's powerful primitives to build an LLM-as-a-judge pipeline that help disambiguate entities in the data.
✅ The DSPy approach is scalable, low-cost and efficient, and is flexible enough to apply to a wide variety of domains and use cases.
Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning
Graph-R1
New RAG framework just dropped!
Combines agents, GraphRAG, and RL.
Here are my notes:
Introduces a novel RAG framework that moves beyond traditional one-shot or chunk-based retrieval by integrating graph-structured knowledge, agentic multi-turn interaction, and RL.
Graph-R1 is an agent that reasons over a knowledge hypergraph environment by iteratively issuing queries and retrieving subgraphs using a multi-step “think-retrieve-rethink-generate” loop.
Unlike prior GraphRAG systems that perform fixed retrieval, Graph-R1 dynamically explores the graph based on evolving agent state.
Retrieval is modeled as a dual-path mechanism: entity-based hyperedge retrieval and direct hyperedge similarity, fused via reciprocal rank aggregation to return semantically rich subgraphs. These are used to ground subsequent reasoning steps.
The agent is trained end-to-end using GRPO with a composite reward that incorporates structural format adherence and answer correctness. Rewards are only granted if reasoning follows the proper format, encouraging interpretable and complete reasoning traces.
On six RAG benchmarks (e.g., HotpotQA, 2WikiMultiHopQA), Graph-R1 achieves state-of-the-art F1 and generation scores, outperforming prior methods including HyperGraphRAG, R1-Searcher, and Search-R1. It shows particularly strong gains on harder, multi-hop datasets and under OOD conditions.
The authors find that Graph-R1’s performance degrades sharply without its three key components: hypergraph construction, multi-turn interaction, and RL.
Ablation study supports that graph-based and multi-turn retrieval improves information density and accuracy, while end-to-end RL bridges the gap between structure and language.
Paper: https://lnkd.in/eGbf4HhX | 15 comments on LinkedIn
Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning
I've spent long, hard years learning how to talk about knowledge graphs and semantics with software engineers who have little training in linguistics. I feel quite fluent at this point, after investing huge amounts of effort into understanding statistics (I was a humanities undergrad) and into unpac
What’s the difference between context engineering and ontology engineering?
What’s the difference between context engineering and ontology engineering?
We hear a lot about “context engineering” these days in AI wonderland. A lot of good thing are being said but it’s worth noting what’s missing.
Yes, context matters. But context without structure is narrative, not knowledge. And if AI is going to scale beyond demos and copilots into systems that reason, track memory, and interoperate across domains… then context alone isn’t enough.
We need ontology engineering.
Here’s the difference:
- Context engineering is about curating inputs: prompts, memory, user instructions, embeddings. It’s the art of framing.
- Ontology engineering is about modeling the world: defining entities, relations, axioms, and constraints that make reasoning possible.
In other words:
Context guides attention. Ontology shapes understanding.
What’s dangerous is that many teams stop at context, assuming that if you feed the right words to an LLM, you’ll get truth, traceability, or decisions you can trust. This is what I call “hallucination of control”.
Ontologies provide what LLMs lack: grounding, consistency, and interoperability, but they are hard to build without the right methods, adapted from the original discipline that started 20+ years ago with the semantic web, now it’s time to work it out for the LLM AI era.
If you’re serious about scaling AI across business processes or mission-critical systems, the real challenge is more than context, it’s shared meaning. And tech alone cannot solve this.
That’s why we need put ontology discussion in the board room, because integrating AI into organizations is much more complicated than just providing the right context in a prompt or a context window.
That’s it for today. More tomorrow!
I’m trying to get back at journaling here every day. 🤙 hope you will find something useful in what I write. | 71 comments on LinkedIn
What’s the difference between context engineering and ontology engineering?
how both OWL and SHACL can be employed during the decision-making phase for AI Agents when using a knowledge graph instead of relying on an LLM that hallucinates
𝙏𝙝𝙤𝙪𝙜𝙝𝙩 𝙛𝙤𝙧 𝙩𝙝𝙚 𝙙𝙖𝙮: I've been mulling over how both OWL and SHACL can be employed during the decision-making phase for AI Agents when using a knowledge graph instead of relying on an LLM that hallucinates. In this way, the LLM can still be used for assessment and sensory feedback, but it augments the graph, not the other way around. OWL and SHACL serve different roles. SHACL is not just a preprocessing validator; it can play an active role in constraining, guiding, or triggering decisions, especially when integrated into AI pipelines. However, OWL is typically more central to inferencing and reasoning tasks.
SHACL can actively participate in decision-making, especially when decisions require data integrity, constraint enforcement, or trigger-based logic. In complex agents, OWL provides the inferencing engine, while SHACL acts as the constraint gatekeeper and occasionally contributes to rule-based decision-making.
For example, an AI agent processes RDF data describing an applicant's skills, degree, and experience. SHACL validates the data's structure, ensuring required fields are present and correctly formatted. OWL reasoning infers that the applicant is qualified for a technical role and matches the profile of a backend developer. SHACL is then used again to check policy compliance. With all checks passed, the applicant is shortlisted, and a follow-up email is triggered.
In AI agent decision-making, OWL and SHACL often work together in complementary ways. SHACL is commonly used as a preprocessing step to validate incoming RDF data. If the data fails validation, it's flagged or excluded, ensuring only clean, structurally sound data reaches the OWL reasoner. In this role, SHACL acts as a gatekeeper.
They can also operate in parallel or in an interleaved manner within a pipeline. As decisions evolve, SHACL shapes may be checked mid-process. Some AI agents even use SHACL as a rule engine—to trigger alerts, detect actionable patterns, or constrain reasoning paths—while OWL continues to handle more complex semantic inferences, such as class hierarchies or property logic.
Finally, SHACL can augment decision-making by confirming whether OWL-inferred actions comply with specific constraints. OWL may infer that “A is a type of B, so do X,” and SHACL then determines whether doing X adheres to a policy or requirement. Because SHACL supports closed-world assumptions (which OWL does not), it plays a valuable role in enforcing policies or compliance rules during decision execution.
Illustrated:
how both OWL and SHACL can be employed during the decision-making phase for AI Agents when using a knowledge graph instead of relying on an LLM that hallucinates
It’s already the end of Sunday — I hope you all had a wonderful week. Mine was exceptionally busy, with the GUG seminar and the upcoming tutorial preparation. I usually take time for a personal…
A Graph-Native Workflow Application using Neo4j/Cypher | Medium
A full working Cypher script that simulates a Tendering System with multiple workflows, AI agent interactions, conversations, approvals, and more — all modeled and executed natively in a Graph.
GraphRAG in Action: A Simple Agent for Know-Your-Customer Investigations | Towards Data Science
This blog post provides a hands-on guide for AI engineers and developers on how to build an initial KYC agent prototype with the OpenAI Agents SDK. We'll explore how to equip our agent with a suite of tools (including MCP Server tools) to uncover and investigate potential fraud patterns.
Transform Claude's Hidden Memory Into Interactive Knowledge Graphs
Transform Claude's Hidden Memory Into Interactive Knowledge Graphs
Universal tool to visualize any Claude user's memory.json in beautiful interactive graphs. Transform your Claude Memory MCP data into stunning interactive visualizations to see how your AI assistant's knowledge connects and evolves over time.
Enterprise teams using Claude lack visibility into how their AI assistant accumulates and organizes institutional knowledge. Claude Memory Viz provides zero-configuration visualization that automatically finds memory files and displays 72 entities with 93 relationships in real-time force-directed layouts. Teams can filter by entity type, search across all data, and explore detailed connections through rich tooltips.
The technical implementation supports Claude's standard NDJSON memory format, automatically detecting and color-coding entity types from personality profiles to technical tools. Node size reflects connection count, while adjustable physics parameters enable optimal spacing for large knowledge graphs. Built with Cytoscape.js for performance optimization.
Built with the philosophy "Solve it once and for all," the tool works for any Claude user with zero configuration. The visualizer automatically searches common memory file locations, provides demo data fallback, and offers clear guidance when files aren't found. Integration requires just git clone and one command execution.
This matters because AI memory has been invisible to users, creating trust and accountability gaps in enterprise AI deployment. When teams can visualize how their AI assistant organizes knowledge, they gain insights into decision-making patterns and can optimize their AI collaboration strategies.
👩💻https://lnkd.in/e__RQh_q | 10 comments on LinkedIn
Transform Claude's Hidden Memory Into Interactive Knowledge Graphs
Building AI Agents with LLMs, RAG, and Knowledge Graphs: A practical guide to autonomous and modern AI agents
𝐁𝐨𝐨𝐤 𝐩𝐫𝐨𝐦𝐨𝐭𝐢𝐨𝐧 𝐛𝐞𝐜𝐚𝐮𝐬𝐞 𝐭𝐡𝐢𝐬 𝐨𝐧𝐞 𝐢𝐬 𝐰𝐨𝐫𝐭𝐡 𝐢𝐭.. 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐚𝐭 𝐢𝐭𝐬 𝐛𝐞𝐬𝐭..
This masterpiece was published by Salvatore Raieli and Gabriele Iuculano, and it is available for orders from today, and it's already a 𝐁𝐞𝐬𝐭𝐬𝐞𝐥𝐥𝐞𝐫!
While many resources focus on LLMs or basic agentic workflows, what makes this book stand out is its deep dive into grounding LLMs with real-world data and action through the powerful combination of 𝘙𝘦𝘵𝘳𝘪𝘦𝘷𝘢𝘭-𝘈𝘶𝘨𝘮𝘦𝘯𝘵𝘦𝘥 𝘎𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘰𝘯 (𝘙𝘈𝘎) 𝘢𝘯𝘥 𝘒𝘯𝘰𝘸𝘭𝘦𝘥𝘨𝘦 𝘎𝘳𝘢𝘱𝘩𝘴.
This isn't just about building Agents; it's about building AI that reasons, retrieves accurate information, and acts autonomously by leveraging structured knowledge alongside advanced LLMs.
The book offers a practical roadmap, packed with concrete Python examples and real-world case studies, guiding you from concept to deployment of intelligent, robust, and hallucination-minimized AI solutions, even orchestrating multi-agent systems.
Order your copy here - https://packt.link/RpzGM
#AI #LLMs #KnowledgeGraphs #AIAgents #RAG #GenerativeAI #MachineLearning
Building Truly Autonomous AI: A Semantic Architecture Approach | LinkedIn
I've been working on autonomous AI systems, and wanted to share some thoughts on what I believe makes them effective. The challenge isn't just making AI that follows instructions well, but creating systems that can reason, and act independently.
LLMs already contain overlapping world models. You just have to ask them right.
Ontologists reply to an LLM output, “That’s not a real ontology—it’s not a formal conceptualization.”
But that’s just the No True Scotsman fallacy dressed up in OWL. Boring. Not growth-oriented. Look forward, angel.
A foundation model is a compression of human knowledge. The real problem isn't that we "lack a conceptualization". The real problem with an FM is that they contain too many. FMs contain conceptualizations—plural. Messy? Sure. But usable.
At Stardog, we’re turning this latent structure into real ontologies using symbolic knowledge distillation. Prompt orchestration → structure extraction → formal encoding. OWL, SHACL, and friends. Shake till mixed. Rinse. Repeat. Secret sauce simmered and reduced.
This isn't theoretical hard. We avoid that. It’s merely engineering hard. We LTF into that!
But the payoff means bootstrapping rich, new ontologies at scale: faster, cheaper, with lineage. It's the intersection of FM latent space, formal ontology, and user intent expressed via CQs. We call it the Symbolic Latent Layer (SLL). Cute eh?
The future of enterprise AI isn’t just documents. It’s distilling structured symbolic knowledge from LLMs and plugging it into agents, workflows, and reasoning engines.
You don’t need a priesthood to get a formal ontology anymore. You need a good prompt and a smarter pipeline and the right EKG platform.
There's a lot more to say about this so I said it at Stardog Labs https://lnkd.in/eY5Sibed | 17 comments on LinkedIn
Graph is the new star schema. Change my mind.
Why? Your agents can't be autonomous unless your structured data is a graph.
It is really very simple.
1️⃣ To act autonomously, an agent must reason across structured data.
Every autonomous decision - human or agent - hinges on a judgment: have I done enough? “Enough" boils down to driving the probability of success over some threshold.
2️⃣ You can’t just point the agent at your structured data store.
Context windows are too small. Schema sprawl is too real.
If you think it works, you probably haven’t tried it.
3️⃣ Agent must first retrieve - with RAG - the right tables, columns, and snippets. Decision making is a retrieval problem before it’s a reasoning problem.
4️⃣ Standard RAG breaks on enterprise metadata.
The corpus is too entity-rich.
Semantic similarity is breaking on enterprise help articles - it won't perform on column descriptions.
5️⃣ To make structured RAG work, you need a graph.
Just like unstructured RAG needed links between articles, structured RAG needs links between tables, fields, and - most importantly - meaning.
Yes, graphs are painful. But so was deep learning—until the return was undeniable. Agents need reasoning over structured data. That makes graphs non-optional. The rest is just engineering.
Let’s stop modeling for reporting—and start modeling for autonomy. | 28 comments on LinkedIn
How can you turn business questions into production-ready agentic knowledge graphs?
❓ How can you turn business questions into production-ready agentic knowledge graphs?
Join Prashanth Rao and Dennis Irorere at the Agentic AI Summit to find out.
Prashanth is an AI Engineer and DevRel lead at Kùzu Inc.—the open-source graph database startup—where he blends NLP, ML, and data engineering to power agentic workflows. Dennis is a Data Engineer at Tripadvisor’s Viator Marketing Technology team and Director of Innovation at GraphGeeks, driving scalable, AI-driven graph solutions for customer growth.
In “Agentic Workflows for Graph RAG: Building Production-Ready Knowledge Graphs,” they’ll guide you through three hands-on lessons:
🔹 From Business Question to Graph Schema – Modeling your domain for downstream agents and LLMs, using live data sources like AskNews.
🔹 From Unstructured Data to Agent-Ready Graphs with BAML – Writing declarative pipelines that reliably extract entities and relationships at scale.
🔹 Agentic Graph RAG in Action – Completing the loop: translating NL queries into Cypher, retrieving graph data, and synthesizing responses—with fallback strategies when matches are missing.
If you’re building internal tools or public-facing AI agents that rely on knowledge graphs, this workshop is for you.
🗓️ Learn more & register free: https://hubs.li/Q03qHnpQ0
#AgenticAI #GraphRAG #KnowledgeGraphs #AgentWorkflows #AIEngineering #ODSC #Kuzu #Tripadvisor
How can you turn business questions into production-ready agentic knowledge graphs?
Find out how to combine a knowledge graph with RAG for GraphRAG. Provide more complete GenAI outputs.
You’ve built a RAG system and grounded it in your own data. Then you ask a complex question that needs to draw from multiple sources. Your heart sinks when the answers you get are vague or plain wrong.
How could this happen?
Traditional vector-only RAG bases its outputs on just the words you use in your prompt. It misses out on valuable context because it pulls from different documents and data structures. Basically, it misses out on the bigger, more connected picture.
Your AI needs a mental model of your data with all its context and nuances. A knowledge graph provides just that by mapping your data as connected entities and relationships. Pair it with RAG to create a GraphRAG architecture to feed your LLM information about dependencies, sequences, hierarchies, and deeper meaning.
Check out The Developer’s Guide to GraphRAG. You’ll learn how to:
Prepare a knowledge graph for GraphRAG
Combine a knowledge graph with native vector search
Implement three GraphRAG retrieval patterns
HippoRAG takes cues from the brain to improve LLM retrieval
HippoRAG is a technique inspired from the interactions between the cortex and hippocampus to improve knowledge retrieval for large language models (LLM).
AI Engineer World's Fair 2025: GraphRAG Track Spotlight
📣 AI Engineer World's Fair 2025: GraphRAG Track Spotlight! 🚀
So grateful to have hosted the GraphRAG Track at the Fair. The sessions were great, highlighting the depth and breadth of graph thinking for AI.
Shoutouts to...
- Mitesh Patel "HybridRAG" as a fusion of graph and vector retrieval designed to master complex data interpretation and specialized terminology for question answering
- Chin Keong Lam "Wisdom Discovery at Scale" using Knowledge Augmented Generation (KAG) in a multi agent system with n8n
- Sam Julien "When Vectors Break Down" carefully explaining how graph-based RAG architecture achieved a whopping 86.31% accuracy for dense enterprise knowledge
- Daniel Chalef "Stop Using RAG as Memory" explored temporally-aware knowledge graphs, built by the open-source Graphiti framework, to provide precise, context-rich memory for agents,
- Ola Mabadeje "Witness the power of Multi-Agent AI & Network Knowledge Graphs" showing dramatic improvements in ticket resolution efficiency and overall execution quality in network operations.
- Thomas Smoker "Beyond Documents"! casually mentioning scraping the entire internet to distill a knowledge graph focused with legal agents
- Mark Bain hosting an excellent Agentic Memory with Knowledge Graphs lunch&learn, with expansive thoughts and demos from Vasilije Markovic Daniel Chalef and Alexander Gilmore
Also, of course, huge congrats to Shawn swyx W and Benjamin Dunphy on an excellent conference. 🎩
#graphrag Neo4j AI Engineer
AI Engineer World's Fair 2025: GraphRAG Track Spotlight
