Knowledge Graph Embeddings as Geometric Operators

Dr. Agus Sudjianto

Dr. Agus Sudjianto works on the connections between geometry, machine learning and model risk. His central idea is that learning is the discovery of geometry, and he pursues it in two ways: in research on interpretable machine learning and in the practical work of making high-stakes models safe to deploy in regulated industries. He advises several AI companies and a university in his field, and consults independently for banks, fintechs and digital-media firms.

This idea shapes how he approaches machine learning. For tabular models, he treats a kernel as a geometry that is learned rather than chosen. That view unifies gradient-boosted trees, Gaussian processes and in-context learning, and it is the subject of his book in preparation, The Learned Kernel. His open-access toolkit MoDeVa brings interpretable model development and validation into one workflow, covering diagnostics, explainability and reliability testing. For generative AI, he applies the same idea to agentic systems through the KnowlytiX framework. Its Geometric Memory System places facts on a learned manifold, so plausibility becomes distance, contradiction becomes measurable tension, reasoning-path consistency becomes a geometric defect and exact numbers are stored without loss. This lets a governed agent sit between symbolic rules and a language model, working as memory, control plane and verifier that can be inspected and audited, and it grounds an agentic harness for testing large language models.

His work builds on more than two decades in quantitative risk. Agus was Executive Vice President and Head of Corporate Model Risk at Wells Fargo, where he chaired the Model Governance Committee and led enterprise model risk management. Earlier he was Director of Analytics and Modeling and Chief Model Risk Officer at Lloyds Banking Group in the UK and Head of Quantitative Risk at Bank of America. His career began in engineering at Ford Motor Company. He holds a PhD in Engineering from Wayne State University and a Master’s in System Design & Management from MIT, holds several U.S. patents and co-authored Design and Modeling for Computer Experiments.

Wing Yan Lau

Wing Yan Lau is the Chief Technology Officer at KnowlytiX and works at the intersection of data systems, knowledge infrastructure and dependable AI. Her focus is the engineering required to turn mathematical ideas into systems that can operate against real enterprise information. In the context of this book, that means connecting knowledge graph embeddings and geometric operators to the practical machinery through which facts are ingested, structured, queried, verified and ultimately made useful to an AI agent.

Her work on the Geometric Memory System spans document ingestion, knowledge-store construction, query infrastructure, verification pathways and the interfaces that connect governed AI systems to enterprise data. She is also a co-author of KnowlytiX research on graph-verified evaluation and structured financial-document retrieval, including work represented in FinStructBench and the company’s broader knowledge and testing platform. That perspective shapes the book’s emphasis on more than mathematical representation alone: an operator must not only express the geometry of a relationship but also be trainable, testable and usable within a complete computational system.

Wing brings more than two decades of database and data-platform engineering experience to this work. She has contributed to core technologies at IBM, SAP and Workday, with responsibilities spanning query optimization, storage systems and execution infrastructure. This background informs how she approaches knowledge graphs: not as abstract collections of triples but as operational systems in which data representation, numerical exactness, retrieval behavior and execution performance must work together.

Her contribution is the implementation discipline that connects theory to practice. She focuses on how documents become structured knowledge, how facts and numeric values retain their fidelity, how graph-backed queries are executed reliably and how verification becomes part of the system rather than an instruction left to a language model. That systems perspective complements the geometric framework developed in this book and helps turn its central ideas into software that researchers and engineers can build, reproduce and run.