# Data science

Each card names a job, the question asked of each row, what the query does with the answers, and the function that asks.

Filter by meaning keeps the rows that are complaints. Category column picks a category for each row, and the query groups by it. Extract fields turns a receipt's free text into amount, date, and vendor columns.

Find duplicates asks whether two rows name the same company and merges the matching pairs. Score and average rates how happy each review is and averages by product. Count topics tags each ticket with its topics and counts them.

Check at load sends incomplete addresses to quarantine. Top N per group keeps the 3 leads in each account likeliest to buy. Second opinion lists the rows where the model and the existing label disagree.

The [tables and frames page](/learn/beatles-bench/data/) shows a real query on Beatles songs. To set one up, start at [Install](/install/).

Ask your data a question, row by row, where it already lives.

On GitHub: [github.com/botassembly/beatles-bench](https://github.com/botassembly/beatles-bench)