Label every row of a table or spreadsheet with a typed question
Add a column whose value Jev fills per row, such as urgency or category, inside DuckDB or a spreadsheet.
Author's figure: Author's figure: about 10 seconds per 1,000 rows (Hamilton Ulmer). Reported by the builder, not measured by us.
The use case
Analysts and operations staff often need one more column: the category of each ticket, the urgency of each request, the topic of each comment. Writing a classifier for a one-off question is not worth it. These builds call Jev once per row from inside DuckDB or a spreadsheet, with a Choice or Score question named in the column header or the SQL. The labelled table can then be grouped and filtered like any other. Someone checks a random sample of rows before the column drives a report or a decision, and keeps the question text with the output so the label is reproducible.
Example workflow
- Load the CSV, Parquet file or table into DuckDB or Google Sheets
- Write the question and allowed labels for the new column
- Call Jev once per row and store label plus confidence
- Check a random sample of 30-50 rows by hand
- Use the column in pivots, filters or dashboards
Tools
- Jev (TypeSafe AI)
- DuckDB
- Google Sheets
Risks and guardrails
- Rows with little text get unreliable labels
- Rerunning with a reworded question changes results
- Large tables cost more than expected if run repeatedly
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Classify rows in DuckDB by Hamilton Ulmer on Made with Jev: madewithjev.com/builds/duckdb-row-classifier
- Classify rows in DuckDB: original post or repo by Hamilton Ulmer: x.com/hamiltonulmer/status/2100370557405667768
- Predictive spreadsheets by Nader Dabit on Made with Jev: madewithjev.com/builds/predictive-spreadsheets
- Predictive spreadsheets: original post or repo by Nader Dabit: x.com/dabit3/status/2100780008193020049