Redash Chatbot Add-on: Agentic RAG for Data Analysis

Ask your dashboards questions in plain English. An add-on for Redash that turns natural-language questions into SQL against existing data sources, so non-analysts get answers without writing queries.

The problem

Redash is powerful but assumes you can write SQL. Business users wait on analysts for even simple questions, and analysts burn time on repetitive ad hoc queries instead of deeper work.

What I built

• An agentic RAG pipeline that grounds the model in the connected schema and existing queries before generating SQL, cutting hallucinated table and column names.
• Natural-language to SQL where the generated query runs against the user's data source and results come back in-chat.
• Schema-aware retrieval so the agent picks the right tables and fields rather than guessing from the question text alone.
• A conversational loop that lets users refine ("only last quarter", "group by region") without starting over.

Stack

Python, LangChain, agentic RAG, Redash integration, LLMs.