Cube x LangChain: Building AI Experiences with LLMs and a Semantic Layer

In an era where deriving insights from data is critical, interest in LLMs is exploding. LangChain has established itself as a core tool for LLM development, while Cube is recognized as a vital component for preventing AI hallucinations and ensuring data accuracy through its semantic layer.

The newly released Cube-LangChain integration is primarily offered as a Document Loader. This loader is used to store embeddings derived from the semantic layer's data models in a vector database. It is then utilized to match free-form inputs, such as natural language queries, with the views and members of the data model.

Additionally, an OpenAI prompt and a Streamlit-based chat demo application are provided. This demo allows users to verify how the tables, columns, aggregations, and filters of SQL queries generated by the LLM align with user input, serving as a starting point for building AI-driven conversational interfaces similar to Delphi.