Caglar Demir
Senior Researcher, Paderborn University
Dr. rer. nat. Caglar Demir is an AI researcher and the Co-founder and CTO of Tentris.
His expertise spans large language models (LLMs), knowledge graph embeddings, and retrieval-augmented generation (RAG).
Currently, he oversees AI product development at Tentris, where he fine-tunes and deploys LLMs ranging from 8 billion to 70 billion parameters, demonstrating a robust capability in enhancing model efficiency and scaling through GPU-optimized CUDA C++ kernels.
In addition to his work at Tentris, Caglar holds a leadership position at Paderborn University as a ML Group Lead Researcher at the Data Science Group, where he directs Ph.D. candidates and software developers in ML research.
His research primarily focuses on creating scalable algorithms for knowledge representation and reasoning, essential for applications in AI-driven inference and retrieval.
His academic journey, marked by a Ph.D. in Computer Science from Paderborn University, centered on developing continuous representations for knowledge graphs, advancing the field of scalable embeddings. His contributions in this area include innovative embedding techniques such as Clifford embeddings and Kronecker decomposition, which offer a more flexible, hardware-agnostic framework for large-scale knowledge graph embeddings, an advancement showcased in projects like "dice-embeddings.
Caglar’s extensive project experience further reflects his commitment to advancing explainable AI (XAI) in industrial settings. His leadership in initiatives like "Rapid Explainable AI for Industrial Facilities" involved designing deep Q-networks integrated with logic programming for industrial diagnostics, addressing critical needs for transparency and efficiency in AI deployment. His recent publications explore topics ranging from universal knowledge graph embeddings to neuro-symbolic learning models that synthesize class expressions, indicating his comprehensive approach to developing interpretable AI models that prioritize balanced trade-offs between interpretability and accuracy.
Through his open-source contributions and practical implementations, Caglar has proven to be a pivotal figure in the intersection of knowledge graph embeddings and RAG, continuously exploring novel ways to improve AI’s reasoning and retrieval capabilities. His work remains instrumental in both academic circles and industry, contributing to the evolution of intelligent systems that better understand and contextualize complex information.
