Data cleaning where the tool suggests the fix
Data preparation is usually the longest stage and the least interesting one. DataLens shortens it by having the assistant read the profile, name the problems it can see, and propose the specific transform for each — which you accept, adjust or ignore.
The transforms
Applied through the interface, recorded as steps, reversible at any point.
Nulls
Fill, drop or flag, with the fill strategy chosen per column and the null rate visible while you choose it.
Duplicates
Detected on the key you nominate or on candidate keys the profiler found, with the collisions shown before anything is removed.
Type casting
Columns that arrived as text but hold dates, numbers or booleans are identified and cast, with the values that would fail listed first.
Renaming
Bulk-rename to a convention, or accept the assistant’s suggestions where a column name says nothing about its contents.
Outliers
Capped, removed or flagged, with the distribution alongside so the threshold is a decision rather than a default.
Joins across datasets
Combine datasets on matched keys, with the match quality reported before the join is committed.
Why every step is recorded
Each transform you apply is written to that dataset’s history — the operation, its parameters, the result and the row count afterwards. The history belongs to the dataset, not to your browser session, so it survives a reload, a sign-out and a switch to another workspace and back.
That matters for two reasons. The obvious one is undo. The less obvious one is explanation: when a number in a report looks wrong, the history is the answer to “what did we do to this data”, and it is available without anyone having to remember.
How a cleaning pass usually goes
Read the findings
The profile has already run. Open the dataset and the quality problems are listed, ranked by how much of the data they touch.
Ask Ari
“Find and fix data quality issues” returns a specific list — this column is 40% null, this one has 12 duplicate keys, this date column is text.
Accept what you agree with
Each proposal is a transform you can apply as-is, adjust, or skip. Nothing is applied without you.
Check the result
Row counts and the profile update after each step, so a transform that quietly destroyed a third of the data is visible immediately.
Keep it
The sequence becomes a repeatable flow when you want the same cleaning applied to the next load of the same source.
Data preparation questions
Can I undo a transform?
Yes. Transforms are applied to a working copy and recorded as an ordered history, so a step can be reversed and the dataset can be reset to the state it arrived in.
Does the AI change my data on its own?
No. It proposes; you apply. Every transform that runs was accepted by a person, and the record of who applied what is kept with the dataset.
Can the same cleaning run automatically next time?
Yes — that is what the orchestration stage is for. A cleaning sequence becomes a flow that runs against the next load of the same source.
See it on your own data
DataLens is in private beta. Bring a file, a database or an API and work through the whole lifecycle in one sitting.
Request beta access