crazycloud/data-analysis-llm-agent

Built on Chainlit with GPT-3.5-turbo and OpenAI function calling, it enables multi-turn conversations where the LLM autonomously decides when to query databases or generate visualizations, with automatic retry logic on database errors. The agent introspects database schema to generate SQL queries and handles errors through iterative refinement—feeding error messages back into conversation context so the LLM can self-correct up to a configurable maximum iterations. Integrates with SQLite and matplotlib for charting, though the README emphasizes key limitations: the LLM can generate syntactically valid but semantically incorrect queries, lacks awareness of actual data distributions, and may struggle with different numeric scales across columns.

No commits in the last 6 months.