Data Neuron is a CLI framework to maintain a semantic/context layer and query DB in natural language. We have been trying to solve the text to sql problem in my company for quite sometime. We have been using a version of semantic layer or the context to get text2sql to some accuracy in our product. The internal framework is UI based and for datawarehouses. I thought I'll create a quick mini version of it (hence the name "neuron", our company is DataBrain) which is a cli tool. Currently supports: SQLite, PostgreSQL, MySQL, MSSQL Works well with Claude 3.5 sonnet which is the default, also has support for OpenAI, LLAMA, Mistral.. The objective is to have a framework or approach so one can create, maintain, improve the semantic layer such that an AI agent can answer questions about one's data. Next set of plans: If one wants to fine tune, transform the context layer to a compatible dataset. And also to create a flask python API to accept incoming questions and return results as json or stream. And to support synthetic query generation based on existing context and to support some form of eval/testing of the generated queries. I have been wanting it to do this cli for quite sometime. This is just a minimal, quickly put together, early version of it. Any feedback would be appreciated! |