New GraphRAG solution aims to give AI 'common sense'

- The core technology, a knowledge graph, acts as a "GPS for AI," creating a structured map of information that connects entities and their relationships. This allows the AI to traverse connections between data points, rather than just finding text with similar keywords, which is the primary method of vector-only RAG. - The 2x reduction in inaccurate answers was measured against the MuSiQue (Multihop Questions via Single-hop Question Composition) benchmark, a dataset specifically designed to test an AI's ability to answer complex questions that require connecting information across multiple sources. - One Graphwise customer, Avalara, a tax software provider, reported improving the accuracy of its AI from 60% with a traditional RAG system to over 90% using the GraphRAG approach. - GraphRAG is particularly effective when the answer to a query is not contained in a single document but must be assembled from the relationships between different pieces of information, a process known as multi-hop reasoning. - The company Graphwise was formed in October 2024 through the merger of two established AI firms: Bulgaria's Ontotext, known for its GraphDB database, and Austria's Semantic Web Company, an expert in knowledge management. - The combined company has over 200 employees with primary locations in New York, Vienna, and Sofia, and is led by President Atanas Kiryakov, formerly of Ontotext. - The solution is designed as a low-code, "production-grade" engine, allowing technical teams to deploy AI workflows with pre-built templates and guardrails for safety and compliance. - Use cases in a business context include enhancing business intelligence by connecting sales data with customer demographics or streamlining knowledge management for complex domains like pharmaceutical research by linking clinical trial data with regulatory guidelines.