shopkeeper-agent
An e-commerce data-warehouse question-answering agent built as a teaching project for learning LangGraph, combining a metadata knowledge base, hybrid retrieval, and staged natural-language-to-SQL generation with execution and streaming results.
shopkeeper-agent answers natural-language questions about e-commerce warehouse data by first building a metadata knowledge base of tables, fields, metrics and field values, then retrieving relevant fields, metrics and value ranges through hybrid search before generating SQL. A vector database handles semantic retrieval of fields and metrics, Elasticsearch handles full-text retrieval of field values, and MySQL stores the complete, authoritative structured metadata.
A LangGraph workflow orchestrates the multi-stage process of retrieval, context assembly, SQL generation, validation, correction and execution, streaming the resulting analysis back to a React frontend over server-sent events. The backend is organized with FastAPI, LangGraph, a repository layer and a client manager, and the project is distributed with a matching free tutorial series and per-chapter git branches so a learner can follow the project's evolution step by step.