Academy
5. Transforming and Cleaning Data
Describe transformations in plain language — filter, derive, aggregate or join — while keeping the original dataset intact.
OUPI Data lets you transform and clean datasets entirely in plain language. Instead of writing formulas or code, you describe what you need — filter rows, derive new columns, aggregate values, or join two datasets — and the AI translates your intent into the actual operation. Crucially, every transformation produces a new view; the original dataset is never modified. This non-destructive approach means you can experiment freely: try different filters, build calculated columns, reshape your data, and always roll back to the source if the result isn't what you expected. Transformations consume credits per operation, and the cost is shown before you confirm.
There are four core transformation types you can request in plain language:
• Filter — Narrow your dataset by conditions: "Show only rows where revenue is above 10 000" or "Keep only Q3 entries." • Derive — Create new columns from existing ones: "Add a column that calculates margin as revenue minus cost." • Aggregate — Summarize data across groups: "Average order value by customer segment." • Join — Combine two datasets on a shared key: "Merge the sales dataset with the product catalog on product ID."
In each case, you write the instruction the way you would ask a colleague. The AI interprets it, runs the operation, and shows you both the result and the underlying logic so you can verify what happened.
Be specific in your transformation requests. Instead of "clean the dates," try "Convert the date column to YYYY-MM-DD format and remove rows where the date is empty." The more precise your plain-language instruction, the more accurate the result — and you can always check the underlying logic the AI shows you.
After running a transformation, save the query behind it. You can re-run it later when the source data is refreshed, or use it as a starting point for further refinements — no need to retype the same instruction twice.
Because transformations never mutate the original dataset, you can chain multiple steps safely. For example: first filter your sales data to the current year, then derive a profit-margin column, then aggregate by product category. Each step creates a refreshed view you can inspect independently. If one step looks wrong, adjust just that instruction — the rest of the chain and the source data remain untouched. This makes iterative data cleaning practical even on large, messy files.
Open the Query screen, pick one of your imported datasets, and type a transformation in plain language — for example: "Filter rows where the amount column is empty, then show the remaining count by category." Review the result and the logic the AI displays, then save the query for later reuse.
Transformations in OUPI Data are described in everyday language — filter, derive, aggregate, or join — and the AI handles execution transparently. The original dataset always stays intact; each operation produces a refreshed view you can inspect, adjust, or discard. Save useful transformations as queries to re-run them on updated data. Every AI-powered operation consumes credits, displayed before you confirm.