Experience with Large Language Models like GPT4 has shown that it is difficult to execute complex multistep solutions. Knowledge graphs are a proven method for guiding chain-of-thought execution by LLMs.
Not all knowledge graphs are equally effective. We have developed a set of structured graphs that capture the context of the problem, organize the content, and guide the LLM to form a well-structured solution.
The talk demonstrates a powerful toolkit that allows domain experts and developers to collaborate on the creation of chain-of-thought solutions. The toolkit contains a low-code diagramming UI and a complementary API designed to integrate with Blazor applications. Once a structured knowledge graph is created using the open-source toolkit, it can be easily integrated into AI-centric solutions, including GPTs, copilots and AI assistants.
Common applications such as proposal generation, product configuration, and legal reasoning are used as examples.