We want to try and evaluate LLMs using Knowledge Graphs

We introduce LLM-ARK, a LLM grounded KG reasoning agent designed to deliver precise and adaptable predictions on KG paths.

Develop an LLM-based Knowledge Graph–augmented Retrieval-Augmented Generation (KG-RAG) system. Implemented advanced graph algorithms to enable deep medical reasoning for fertility-related use cases

A knowledge-graph–driven DDx system that combines Bayesian reasoning, adaptive question selection, and natural language understanding. The system grounds patient symptoms into a medical KG, ranks candidate diseases using probabilistic traversal, and then uses an LLM to generate explanations.

No description available

No description available

We introduce LLM-ARK, a LLM grounded KG reasoning agent designed to deliver precise and adaptable predictions on KG paths.

No description available

AI-powered enrichment pipeline for hc-enterprise-kg — LLM reasoning, web search grounding, and GraphGuard quality contracts

Developed a full pipeline for event/causal extraction, ATOMIC triple generation/translation, KG integration, R-GCN-based commonsense inference, and subsequent LLM reasoning enhancement using KG-derived embeddings

Leibniz Lab PyReason Assignment combining a knowledge graph (KG) + an LLM + logical reasoning to infer new facts that were never explicitly stated.

KG-Hopper is a novel Reinforcement Learning (RL) framework that empowers compact open LLMs with the ability to perform integrated multi-hop reasoning within a single inference round.

A Neuro-Symbolic KG-RAG Diagnostic Engine. Achieves <500ms retrieval latency by fusing vector search (Qdrant) with structured medical ontologies (Neo4j/PrimeKG) to ground LLM reasoning in verified clinical protocols.

This initiative builds a Knowledge Graph (KG) with a Large Language Model (LLM) for advanced search and reasoning. The KG integrates entities like companies, skills, roles, and salaries using unified taxonomies. It enables plain language queries, insightful reports, and future capabilities like predictions and recommendations.

A production-grade multi-modal RAG stack combining OCR, LLM + VLM reasoning, hybrid retrieval, and GraphRAG-inspired KG/embedding context selection. Includes reward-style fine-tuning + RL post-training (RLHF) under low-data constraints, plus automated eval suite and scalable PyTorch training pipelines.