Building Knowledge Graphs to Enable Agentic AI with Context

This slide deck accompanies a talk presented at the Machine Learning Summit 2025 organized by Pcakt. I focus on a key idea: in ecosystems of intelligent agents, both human and artificial, performance depends not only on individual intelligence but on the ability to interoperate and create higher-order functionality aligned with organizational goals

Autonomy is not enough. The real challenge is semantic interoperability, making it possible for agents to share context, interpret meaning consistently, and coordinate effectively. Achieving this requires a shared information architecture with stable, machine-readable semantics

I present an approach for building such a semantic layer using a knowledge graph that integrates data products, data contracts, and ontologies. I show how this foundation enables scalable, reliable agentic systems, including LLM-powered agents capable of planning and coordination

I also highlight practical design patterns and techniques to ensure long-term sustainability and operational robustness. This talk demonstrates how a well-structured semantic layer can turn an information architecture into a strategic enabler, helping organizations unlock the full potential of both human and artificial intelligence

Transcript

1. 1. Language agents need proper context to perform tasks reliably. 2. Managing retrieval and learning processes is key to providing the right context at each step. 3. An enterprise knowledge graph facilitates agents' retrieval and learning processes. Building Knowledge Graphs to Enable Agentic AI with Context Andrea Gioia Starting Now CTO at Quantyca Cofounder at Blindata.io Author at Packt
2. Building Knowledge Graphs to Enable Agentic AI with Context
3. Andrea Gioia CTO at Quantyca Cofounder at Blindata.io Author at Packt About ANDREA GIOIA • Partner and CTO at Quantyca, a consulting firm specializing in data management. • Co-founder of Blindata.io, a SaaS platform for data governance and compliance. • Over 20 years of experience leading complex data projects across sectors such as banking, utilities, retail, and industry. • Current responsibilities include advising clients on data strategy with a focus on organizational and change management. • Active contributor to the data community through speaking engagements, publications, and the book Managing Data as a Product. • Member of DAMA and, since 2023, part of the scientific committee of the DAMA Italian Chapter.
4. Agenda 01 Context Engineering 02 Knowledge Mesh 03 Enterprise Knowledge Graph
5. Section 01 Context Engineering
6. The Rise Of Language Agents > Context Engineering Rule Engine Symbolic AI Agent AI Winter 1960 1980 2010 2022 2024 LLM Language AI Agent ENVIRONMENT ✅ Scalable: Design and train is outsourced, adaptation require low manual efforts ✅ Practical: Task-independent, easy to generalize ❌ Low Domain Accuracy: May hallucinate in domain specific contexts ❌ Low Transparency: Opaque decision-making, hard to interpret or trace reasoning Examples: Voyager, Manus, DeepSearch, Operator ❌ Not Scalable: Intensive manual efforts to design and train, adaptation is hard even for experts ❌ Not Practical: Task-specific, hard to generalize ✅ High Domain Accuracy: Performs exceptionally well in narrowly defined, rule-based domains ✅ High Transparency: Reasoning process is traceable and decisions are explainable Examples: Deep Blue, Watson, MYCIN, DENDRAL, XCON 🤓 🤯 🤒 ENVIRONMENT Pg: 6
7. > Mind the Gap Building an impressive AI Demo has never been easier ✅ turning the demo in a reliable production-ready AI Product remains challenging ❌ LLM ENV ENV LLM LLM LLM LLM DEMO PRODUCTION … BUT … Context Engineering Pg: 7
8. > Context is all you need Most of the time when an agent is not performing reliably the underlying cause is not that the model is not intelligent enough, but rather that context has not been communicated properly to the model. “The single biggest problem in communication is the illusion that it has taken place.” George Bernard Shaw 📌 Communication is all you need by LangChain Context Engineering Pg: 8
9. > Why large context windows are not enough Large context means slow responses. Scanning all the tokens on every request takes time. LATENCY Large context windows are expensive. More tokens means more money spent. COSTS Large context can easily become poisoned, distracting, confusing, or conflicting. ACCURACY 1M 📌 Will Growing Context Window Size Kill RAG? by Vishvambhar D. 5M 15M Claude 4 & GPT-04 Gemini 2.5 Ultra Llama Horizon+ Context Engineering Pg: 9
10. > Context Engineering Context Window Context Engineering Memory State Knowledge Prompts Retrieval Processes 📌 The rise of "context engineering" by LangChain Learning Processes Context Engineering Pg: 10 Context engineering is the art and science of filling the context window with just the right information at each step of an agent’s trajectory. Andrej Karpathy
11. > Memory types Know How Know Why Know That Know When Know What 📌 Cognitive Architecture For Large Language Agents by Y. Shunyu et al. CoALA is a framework to organize existing agents and plan future developments. Under the CoALA framework, language agents explicitly organize information into multiple memory modules, each containing a different form of information. These include short-term working memory and several long-term memories: episodic, semantic, and procedural. Context Engineering Pg: 11
12. > Information Architecture Context Engineering Short term memory Reasoning Retrieval Learning Tools ENVIRONMENT Episodic memory Working memory Semantic memory DATA KNOWLEDGE INFORMATION Long term memory DATA Understand KNOWLEDGE INTELLIGENCE Act Comprehend INFORMATION Processing Cognizing Reasoning Sensing Collect Pg: 12
13. > Pg: 13 Adhocracy creates an agentic mess Context Engineering IA IA AI AI IA Customers have opened N tickets this week Customers have purchased N items this week We have shipped N items to customers this week When they talk about customers, are they referring to exactly the same thing? Siloed ad-hoc Context (aka Information Architecture) RAG AI
14. > Pg: 14 Enterprise Information Architecture Context Engineering IA Enterprise Customers have opened N tickets this week Customers have purchased N items this week We have shipped N items to customers this week AI RAG RAG RAG RAG IA IA IA IA
15. Section 02 Enterprise Knowledge Graph
16. > Pg: 16 Enterprise Knowledge Graph EKG Enterprise Ontology DATA KNOWLEDGE INFORMATION Processing Cognizing Reasoning Sensing Upper ontology Domain ontology Physical Data Subdomain ontologies An enterprise knowledge graph is an elegant way to represent the entire information architecture in a single model that is • Shared • Formal • Computable
17. > Pg: 17 The power of modality EKG / Data Plane WTF! 👍 Monolithic Data Solution Modular Data Solution
18. > Pg: 18 Data products EKG / Data Plane Accuracy Relevance Reusability Composability VALUE Identify and maintain Share and multiply INTERFACES DATA METADATA CODE INFRASTRUCTURE Data Pipeline Storage Compute
19. > Pg: 19 Metadata management shift-left EKG / Information Plane INTERFACES DATA METADATA CODE INFRASTRUCTURE Data Pipeline Storage Compute Promises Expectations Obligations Populates Accepts & consumes Consumers Data Contract Federated Governance Standardizes Self-Serve Platform Supports INTERFACES DATA METADA TA CODE INFRASTRUCTURE Data Pipeline Storage Comput e Syntactic & technical interoperability Producer
20. > Pg: 20 Managing Data as a Product EKG / Data & Information Plane DATA PRODUCT Discovery ports Output ports Data Plane Information Plane Data Product Team
21. > Pg: 21 Lost in translation EKG / Knowledge Plane Different data assets are composable when they are interoperable at the following levels: 1. Technological 2. Syntactic 3. Semantic Semantic interoperability is a major challenge.
22. > Pg: 22 Knowledge as a first-class citizen EKG / Knowledge Plane I know the concept of “tree” because I know that … … a tree is not a mineral … a tree is a plant … a tree is not a bush … plus other facts that I have learned through experience or reasoning … a tree has a trunk Mineral Natural thing Plant Bush Tree Trunk :type of :type of :type of :type of :has Implicit & Personal Knowledge Model Explicit & Shared Knowledge Model Formalized into Grounded on
23. Pg: 23 Semantic linking EKG / Knowledge Plane > infra API Data Contract Ports Exposes Internal Components Schema Links to Data Product Data Product Descriptor Enterprise Ontology Information Plane Knowledge Plane Data Plane Data Product Team Modelling Teams Dati Apps Search INTO data Search FOR data
24. Section 03 Knowledge Mesh
25. > Pg: 25 Federated conceptual modeling EKG / Knowledge Mesh Self-serve Platform X as a Product X Domain Ownership Computational Governance Knowledge Mesh Data Mesh Socio Technical
26. > Pg: 26 Knowledge as a Product EKG / Knowledge Mesh Enterprise Ontology Upper Ontology Domain Ontologies Subdomain Ontologies Ontology Lifecycle
27. > Pg: 27 Knowledge as a Product EKG / Knowledge Mesh Market ing Sales EMEA Sales Nordic s Operat ions Business Domains Knowledge Domains Ontology Factual Data Custome r Product Order Knowledge Domain Owner Knowledge Domain Owner Knowledge Domain Owner Business Domain Owner Business Domain Owner Business Domain Owner Business Domain Owner A federated modeling team composed of representatives from each business domain manage the definition of the enterprise ontology. This team can further organize by dividing responsibilities based on knowledge domains.
28. > Pg: 28 Computational Governance EKG / Knowledge Mesh Domain Ontologies Federated Governance Team Self-serve Platform GLOBAL POLICIES Defines Enforces Conforms to
29. DATA INFORMATION KNOWLEDGE Data Product Developer Platform Data Product Catalog Knowledge Product Developer Platform > Pg: 29 Self-serve Platform EKG / Knowledge Mesh
30. > Pg: 30 Incremental & value-driven approach EKG / Knowledge Mesh Ontology Data products Knowledge Graph Linking Knowledge Plane Concepts + Relationships Information Plane Data + Metadata Data Management Solution Deploy Deploy Iterate Business Cases Modeling Team Data Product Team Business Analysis Knowledge Modeling Data Product(s) Implementation
31. > Pg: 31 Final takeaway EKG / Knowledge Mesh There is no reliable AI without a reliable IA Artificial Intelligence Information Architecture OUTSOURCE INTELLIGENCE INSOURCE KNOWLEDGE
32. Thank You