Semantic Web Company and Ontotext merge to form Graphwise

Vienna-based Semantic Web Company (SWC) is combining with Bulgaria-headquartered Ontotext to become Graphwise to better meet the retrieval-augmented generation (RAG) needs of generative AI by providing Graph RAG.

The two say SWC brings expertise in knowledge engineering, semantic AI, and intelligent document processing, while Ontotext brings a versatile graph database engine and state-of-the-art AI models for linking and unifying information at scale. Graphwise now has “the most comprehensive knowledge graph management platform, which includes complete multi-modal data support – unstructured, semi-structured, and structured data.”

Atanas Kiryakov

Graphwise president Atanas Kiryakov stated: “Knowledge graphs are like a GPS for AI and large language models (LLMs). They guide AI models with precision and context to ensure trustworthy, explainable outputs. Just as a GPS system provides accurate routes and prevents wrong turns, knowledge graphs steer AI models in the right direction by organizing and linking data in meaningful ways. The ability to do this has never been so important, as businesses grapple with multiple AI technologies.” 

The company says that by using knowledge graphs, enterprises get more accurate, context-rich insights from their data, which is essential as they look to adopt AI to drive decision-making.

As we wrote here, knowledge graphs store and model relationships between entities (events, objects, concepts, or situations). There are head and tail entities, with a “triple” referring to a head entity + relationship + tail entity. Such triples can be linked and the relationships between them form semantic knowledge.

Generative AI LLMs use semantic search based on finding vectors (encoded tokens describing abstracted aspects of text, audio, image or video data) to generate responses to users’ requests, based on training with general unstructured data sets. RAG makes their responses more accurate by bringing in private datasets. Graph RAG refers to making LLM responses more accurate and better informed still by using knowledge graphs to provide contextual information missing from generic RAG.

For example, an LLM can base predictions and inferences on the factual relationships represented in a knowledge graph. If an LLM is asked to generate information about a prominent person, it could access a knowledge graph to get specific data points, such as date of birth, qualifications, employment status, awards, etc. and generate a more accurate response.

Andreas Blumauer

LLMs need additional training to use knowledge graphs. Pre-training can involve the LLM learning to recognize entities and relationships in natural language, using entity embeddings from the knowledge graph. For example, entities like “Musk” and “electric cars” may be represented as knowledge graph nodes, and the LLM learns how they are connected through graph-based representations.

A knowledge graph’s structure of entities and relationships can be vectorized into knowledge graph embeddings, with entities becoming nodes and relationships becoming edges in a continuous vector space. For example, a specific model called TransE produces knowledge base embeddings for relationships. A method called DistMult represents entities and relations as embedding vectors in semantic space and predicts the links between them. 

Martin Kaltenbock

SWC was founded in 2004 in Vienna, Austria, by CEO Andreas Blumauer and CFO Martin Kaltenbock. Blumauer was also the CEO of knowledge graph company punkt. netServices from 1998 onward. SWC was spun off from punkt. netServices in 2004 and then merged back in 2011. Before this, Blumauer was a lecturer in Knowledge Management Systems at FHWien der WKW, Austria’s main university of applied sciences for management and communication. Kaltenbock was the managing director of punkt. netServices from 2000 to 2011. Blumauer is now the SVP for growth at Graphwise. Kaltenbock’s Graphwise role has not yet been revealed.

SWC raised an undisclosed amount of funding in a 2014 seed round. Its main product is PoolParty, semantic information management software used by more than 180 customers. SWC says it enables Graph RAG applications for generative AI.

Ontotext was founded in 2008 by CEO Atanas Kiryakov in Sofia, Bulgaria, when it raised an unrevealed funding amount in an A-round. He was a software engineer at the Sirma Group from 1993 onward, becoming its owner in 1996 and head of its Ontotext Lab in 2000. Ontotext has a GraphDB database engine product that delivers knowledge graphs and is a graph RAG product.

Kiryakov is a co-founder and member of the board of the Linked Data Benchmarking Council (LDBC), an association of graph database vendors.

Graphwise has more than 200 employees worldwide and offices in Boston, Switzerland (Basel), London, New York, Sofia, and Vienna.