Dataset Quality Filter: check photos, video and text before your AI learns from them
Example request“Count the cars in my photos, and flag photos that are blurry, too dark or copies.”
Problems we flag
- Blurry photos
- Too dark or too bright
- Copies of the same photo
- Broken files that will not open
What is inside (Content)
- We count the things you ask for, from our approved list
- We count faces, but never find out who someone is
- Outlines for supported objects
Planned file types: JPG, PNG, WebP · up to 30 megapixels (most phone and camera photos)
Example request“Count the cars in my dashcam videos, one frame per second.”
Problems we flag
- The picture freezes
- Parts that are too dark
- A dirty or blurry lens
You get the exact time each problem starts, so you can keep the good parts.
What is inside (Content)
- We count the things you ask for, frame by frame
- People are only counted, never identified
Planned file types: MP4, MOV, MKV, WebM and other common formats · up to 8K
Example request“Find personal details and secret keys in our support emails, and flag copies.”
Problems we flag
- Copies of the same text
- Texts that are almost the same
What is inside (Content)
- Email addresses and phone numbers
- ID numbers
- Secret keys and access codes that should not be shared
Planned file types: TXT, CSV, JSON, JSONL, Markdown, HTML, PDF, Parquet
What you get back
- Fits your request
- Has what you don’t want
- For you to check
- Could not check (you don’t pay for these)
Every file gets a reason. We never delete or change your files.
You also get a datasheet, a short summary of what your files hold, and a sealed record that proves nothing was changed (planned).
Good to know
- Dataset Quality Filter is still in development. Uploads and payments are not open yet.
- Checks can miss things or flag something by mistake. You always have the last word.
- The checks do not make you compliant with the GDPR or the EU AI Act. They help you see what is in your data.
- Made for anyone building or training their own AI in Europe, from a two-person start-up to a large team. More in our guides. For a price, contact us.
Want to try it on your own files?
We are looking for our first pilot teams.
Talk to us about a pilot