A deep dive into the evolution of semantic AI and knowledge graphs, and how Graphwise is redefining enterprise data strategy through scalable Graph AI.
Atanas, with your extensive background in semantic databases and knowledge graphs, can you share the journey that led you to founding Ontotext and later becoming the President of Graphwise?
In the late 90s, I was involved in research and teaching AI at Sofia University. This was part of my PhD program which was running in parallel to my “day job” as a manager at Sirma which was one of the first software companies in Bulgaria. In Q2 of 2000, I went to the University of Tuebingen to dedicate 4 months to my PhD thesis. There, I developed some ideas about how ontologies and text analysis can help search. In July, I proposed to my partners at Sirma that we start an R&D lab around these ideas, and they approved a budget of $15,000. This is how Ontotext began, and I never finished my PhD.
What inspired you to focus on knowledge graphs and semantic AI technologies, and how have these fields evolved since you began your career?
In the late 90s AI was in decline – expert systems went out of fashion by 1995 and there was not enough data and computing power to get good results with neural networks. Fuzzy logic was used in washing machines without calling it AI. There was even a saying “If it works, it’s not AI.” At the same time the Internet was booming and Sir Tim Berners-Lee, the inventor of WWW, developed the idea for the Semantic Web, which is a web of data that is not only good for human consumption but also easy for machines to use as if it is in a database. Ontologies and graph data representation were at the core of this vision and this is how AI legacy powered something new.
Around 2012 semantic technologies reached the disillusionment phase. Google was the one to coin the term Knowledge Graph and to educate people on how it helps search engines in a big way. Ironically, this is what we have been doing at Ontotext for more than 10 years. Now that the market understood the concept and benefits, we started developing enterprise knowledge graphs – a tailor-made variant of what Google uses in w2c setting.
By 2019, graph databases had matured and border adoption began. At the same time, the rise of transformers and later large language models (LLMs) helped lower the cost of text analysis and made it easier for organizations to populate knowledge graphs from unstructured content.
Graphwise emerged from the merger of Ontotext and Semantic Web Company. Can you talk about the vision behind this merger and the synergies it has created?
Ontotext and Semantic Web Company (SWC) started a strategic partnership in 2020. This allowed us to get to know each other, make the initial round of integration of the company’s technologies and identify the synergies of a potential merger.
Our shared vision is to bring confidence to AI by helping enterprises make their data and content AI-ready. The merger gave us the critical mass needed to compete with the big vendors which are entering the graph space urged by the increased demand for knowledge graphs related to GenAI. As a result of integrating the two companies, Graphwise has created the first end-to-end Graph AI platform and tools combined with a team of experts who can deliver our solutions worldwide.
In your view, why are knowledge graphs becoming increasingly essential for enterprises investing in AI?
Knowledge graphs help enterprises get their data AI-ready in several ways. First, they connect data across silos, guarantee data consistency, and ensure that the meaning of the data is appropriately aligned and unified to reduce data quality issues and misinterpretation. Second, proprietary data gets enriched with domain knowledge which is the context that GenAI needs to properly interpret the data. For example, when the word “Paris” is in the text presented it can help determine if this Paris is in reference to the city in France, in Texas or the person Paris Hilton. Similarly, when the abbreviation “NASA” is used, it recognizes it as a US government agency. These are things that people know, but databases don’t understand and LLMs are shaky at. Finally, semantic graph databases offer automatic reasoning that uncovers new relationships and surfaces implicit knowledge. All these factors together increase the performance of GenAI and make it much more trustworthy.
How does Graphwise ensure that its knowledge graph solutions are scalable and trustworthy for large enterprises?
Speaking about scalability, people often think of data size and sharding – a database architecture pattern related to horizontal partitioning – which is everything the big data trend was about. For a database engine, there are several additional dimensions of scalability related to the efficient handling of a big number of concurrent queries, complex analytical queries, and frequent update transactions. Graphwise’s GraphDB solution has been proven to balance all these requirements well.
GraphDB was developed by Ontotext since 2004 with the goal of providing predictable performance across a wide range of workloads. Audited benchmarks prove it is the most versatile graph database that efficiently handles graph analytics and semantic metadata management. Unlike other offerings, GraphDB is the only engine passing both LDBC Semantic Publishing and Social Network Benchmarks. There is also a long list of enterprise customers that rely on business-critical deployments of GraphDB, spanning across multiple data centers and regions.
Scaling up is also about having processes and tools for efficient and sustainable development of knowledge graphs. For instance, many people say “I can craft a taxonomy in Notepad or Excel.” Yes, they can, but if you want a team that collaboratively maintains and develops a taxonomy of thousands of concepts, which is then used by thousands of people and applications, then you need a proper knowledge modeling tool. This is what SWC’s PoolParty has been optimized for for 20 years.
Could you explain the critical role of semantic AI technologies in helping organizations maximize the ROI from their AI investments?
Knowledge graphs allow for easier data discovery and reuse. The Semantic Web standards (RDF and related) are proven to be the best paradigm for data publishing and sharing and for development of distributed data architectures. Data preparation work done on one AI project can be reused in other projects rather than thrown away. The latter is what happens if the so-called property graphs (LPG) are used instead of RDF. To understand the savings scale, consider that about 80% of the work on an AI project is data preparation.
What are some of the most common challenges companies face when integrating knowledge graph infrastructure, and how does Graphwise address these issues?
Many enterprises struggle to find architects and managers who understand the technology enough to plan initiatives optimally and manage risk. Graphwise can offer an end-to-end solution that takes care of the entire process – from scoping to planning, delivery, knowledge transfer and handover. To scale this properly, we created a global ecosystem of partners composed of local consultants, value-added resellers and global system integrators who can take care of this for our customers.
Another challenge for many organizations is staffing projects with people skilled in knowledge modeling and data engineering. Beyond the obvious option to outsource this work to one of our partners, we also offer implementation approaches that reduce the need for this work. For instance, we help enterprises use existing ontologies and schemas, such as Schema.org. Beyond the cost savings from avoiding the need to design your own ontology, all major LLMs already “know” these schemas, so, no finetuning is needed to implement Graph RAG. Another example is PoolParty’s Taxonomy Adviser, which automatically bootstraps or extends a taxonomy form documents.
Can you provide an example of a successful implementation of Graphwise’s technology in a large enterprise and the outcomes achieved?
Yes. One of the biggest banks in the USA uses GraphDB to bring together in a graph monitoring information from several of their most critical retail banking IT systems. Inference is used to further reveal multi-hop relationships and enable more robust analytics. One of the outcomes is that they can substantially reduce the time for root cause analysis in case there is a problem. The system also allows them to automate the what-if analysis performed to discover potential vulnerabilities and improve operational resilience.
How does Graphwise differentiate itself in a competitive market of knowledge graph and semantic AI providers?
We are unique in several ways. First, we have GraphDB – the most versatile graph database engine. Second, we have PoolParty – the market-leading knowledge management tool. Third, we are the only vendor that offers an end-to-end platform for data, knowledge and content management. Finally, in recent years we have been recognized as the established leader in combining graph technology with other AI techniques. We offer a rich toolbox that includes a range of AI models complementary to the LLMs, and retrieval methods. This allows AI architects to find the best balance between performance and cost for each specific project.
Finally, where do you see the future of knowledge graphs and semantic AI heading, and what role do you envision Graphwise playing in that future?
One trend is that knowledge graph technology will penetrate the mainstream market within a couple of years. There will be a growing number of applications where graph technology will be embedded and users will benefit from it without knowing it’s under the hood.
Another trend is that there will be a growing number of vertical niche business solutions, built on knowledge graphs. One example is our Target Discovery solution, which accelerates the drug development process. Another example is rECOmentor – a solution that reduces the efforts and risks in ESG reporting. Expect to see a growing number of solutions developed by a spectrum of vendors – from startups to GSIs and remember this is only the beginning.
A quote or advice from the author: Graphwise’s vision is to bring confidence to AI by helping enterprises make their data and content AI-ready. The merger between Ontotext and Semantic Web Company gave us the critical mass needed to compete with the big vendors which are entering the graph space urged by the increased demand for knowledge graphs related to GenAI. As a result of integrating the two companies, Graphwise has created the first end-to-end Graph AI platform and tools combined with a team of experts who can deliver our solutions worldwide.
Atanas Kiryakov
President of Graphwise
Atanas Kiryakov is President of Graphwise – the leading Graph AI provider after the recent merger between Semantic Web Company and Ontotext. For more information visit the company at https://graphwise.ai/ or connect on LinkedIn.
Explore AITechPark for the latest advancements in AI, IOT, Cybersecurity, AITech News, and insightful updates from industry experts!