October 2025 Summaries

Large enterprises often face challenges with large language models (LLMs) managing vast, complex data graphs, but GraphRAG offers a solution by teaching models to reason through structure rather than feeding them more data. The approach focuses on using enterprise graph schemas to help LLMs understand entity connections, leveraging Memgraph's fine-grained access control to filter schemas and reduce noise, thereby enhancing focus and accuracy. By granting users like the hypothetical "Jim," a new developer, access to only relevant parts of the schema, LLMs can reason effectively within a scoped context, preventing the distraction of vast, irrelevant information. This method enables the creation of LLMs that are not only smarter and more efficient but also better equipped to deliver precise, domain-specific insights in large organizational settings. Ultimately, fine-grained access control helps LLMs become focused experts by providing the right context, ensuring that they reason accurately without being overwhelmed by extraneous details.

Retrieval-augmented generation (RAG) and GraphRAG are both designed to enhance large language models (LLMs) by providing them with more contextually rich information than they can access from their training data alone. RAG achieves this by retrieving semantically relevant information from external sources, allowing models to generate responses grounded in current facts, making it suitable for tasks involving unstructured text like FAQs and customer support. However, RAG lacks the capability to understand relationships between pieces of information. GraphRAG addresses this limitation by using a knowledge graph structure that captures not only the semantic relevance of information but also the relationships and reasoning behind it, making it ideal for complex queries requiring multi-hop reasoning or relational understanding, such as supply chain analysis and healthcare intelligence. GraphRAG builds upon RAG's retrieval foundation, offering a more comprehensive approach by integrating both semantic matches and graph reasoning while maintaining flexibility through various retrieval methods. Together, they represent layers of the same ecosystem, with GraphRAG evolving the concept of retrieval by adding a dimension of understanding through connected context.

GraphRAG is an innovative approach that enhances the capabilities of large language models (LLMs) by connecting context through knowledge graphs, which organize information as structured representations of entities and their relationships. Unlike traditional LLMs that rely on pattern matching, GraphRAG integrates semantic retrieval and graph reasoning to provide a more comprehensive understanding of data, allowing models to reason across relationships and not just retrieve information based on keyword similarity. This method is particularly effective in domains where the accuracy and interconnections of data are crucial, such as healthcare and logistics, enabling systems to adapt to dynamic data and support informed decision-making. Platforms like Memgraph implement GraphRAG by supporting various search strategies and graph analytics, facilitating real-time data processing and continuous learning from new information. This approach transforms fragmented data into meaningful knowledge, offering a scalable solution that aligns with the growing complexity of organizational data needs.

Retrieval Augmented Generation (RAG) is a method that enhances large language models (LLMs) by allowing them to access external data, thus overcoming their inherent limitation of not knowing information beyond their training data. RAG works by converting both data and queries into embeddings—high-dimensional vectors encoding semantic meaning—stored in a vector database, which the LLM can search to find contextually relevant information when generating responses. This approach enables LLMs to provide more accurate and grounded answers by using fresh data without requiring model retraining, making it especially useful for unstructured data and applications like enterprise chatbots and document Q&A systems. However, RAG has limitations, such as losing hierarchical context and struggling with complex domain relationships, which can lead to errors in reasoning and relevance. Despite these challenges, RAG remains a powerful tool for bridging the gap between LLMs and real-world data, forming the backbone of reliable semantic search systems.

GraphRAG (Graph-based Retrieval-Augmented Generation) leverages graph databases to transform fragmented healthcare data into cohesive knowledge graphs, enhancing the ability to analyze and understand complex relationships within the data. This approach, facilitated by Memgraph, allows for the integration and real-time interaction of disparate data sources such as patient histories, lab results, and insurance data, overcoming the limitations of traditional databases that struggle with contextual relationships. By employing graph algorithms for real-time analytics, GraphRAG reveals insights and patterns buried within healthcare systems, enabling efficient care delivery, research collaboration, and operational cost reduction. The technology excels in modeling relationships and provides context-aware insights, enabling healthcare teams to proactively explore care pathways and research opportunities. As a result, GraphRAG offers a significant advancement in healthcare information management by turning data silos into dynamic knowledge networks that enhance discovery and innovation in the field.

In the recent Memgraph Community Call, Vasilije Markovic, founder of Cognee, discussed the limitations of current retrieval-augmented generation (RAG) systems and introduced Cognee as a memory system for AI agents that enhances reliability through a graph-powered memory layer. Traditional large language models struggle with contextual memory, leading to inefficiencies in production systems. Cognee addresses this by offering persistent, adaptive memory using embeddings combined with graph-based extractions, improving recall accuracy to around 90% and making AI assistants more reliable for decision-making. The system is implemented as a Python SDK, gaining traction in open-source projects with significant early adoption. The call highlighted the shortcomings of RAG, such as operational challenges and mismatches in semantic similarity, and demonstrated how Cognee's approach overcomes these by utilizing knowledge graphs for context-rich entity representation. Early iterations of Cognee showed promise but required further enhancements in parallelism, customization, and performance. The current architecture integrates vectors, graphs, and reasoning, providing a developer-friendly stack that is both modular and scalable. Future developments focus on expanding features to support real-world deployments, with ongoing efforts to improve API usability and scalability to larger datasets.

Clinical trials, essential for developing new medical treatments, often face delays primarily due to challenges in patient recruitment and retention. Graph-powered applications offer solutions by creating Patient Journey Knowledge Graphs (PJKGs) that integrate patients' medical histories, demographics, and treatment data into a comprehensive network, revealing hidden patterns and relationships. This graph-based approach enhances the recruitment process, transforming it from a cumbersome manual task into a precise, data-driven operation, facilitated by algorithms such as Community Detection, Link Prediction, and Betweenness Centrality. These algorithms improve trial design and execution by accurately identifying patient cohorts, predicting enrollment likelihood, and optimizing resource allocation. In trials for rare diseases, graphs swiftly identify connected patient communities, overcoming the limitations of traditional, fragmented datasets. The visual mapping of these connections aids decision-makers in streamlining recruitment and reducing trial dropouts, ultimately ensuring trials remain on schedule and yield more predictable outcomes. Beyond clinical trials, graph technology holds potential for broader applications in personalized patient care, drug discovery, and healthcare analytics, offering transformative insights into the relationships within healthcare data.