TidyTuesday Agent
An AI agent that autonomously answers data analysis questions about TidyTuesday datasets. Uses a 5-step workflow: understand the question, load data into DuckDB, run SQL queries, interpret results, and format answers. 61 tests, domain-aware question interpretation.
View project →GitHub: prahlaadr/tidytuesday-agent
Why I Built This
Wanted to build an AI agent that could take a plain-English question about a dataset and autonomously figure out the answer — loading data, writing SQL, interpreting results — without human guidance.
How It Works
5-step workflow: understand the question → load CSV into DuckDB → generate and run SQL queries → interpret results → format a human-readable answer. Domain-aware question interpretation maps natural language to SQL operations. DuckDB for fast analytical queries. 61 tests cover edge cases and complex queries.
Notable Technical Details
- Domain-aware natural language to SQL interpretation
- DuckDB for fast in-process analytical queries
- 61 test cases covering edge cases and complex queries
- Autonomous multi-step reasoning pipeline
- Handles ambiguous and open-ended questions gracefully