Soon live
Dataset Quality FilterFilter first. Train second.
We check every photo, video and text before your AI learns from it. You decide what stays. Built for teams building AI in Europe.
supportgate
Dataset Quality FilterChecks for photos, video and text. Soon live.
- ImageBlurry, dark or broken photos and copies. Counts the things you ask for.
- VideoFrozen or dark parts and a dirty lens, with the time. Counts the things you ask for.
- TextCopies, personal details and secret keys
SupportGate is a member of the NVIDIA Inception program.
© 2025 NVIDIA, the NVIDIA logo, and NVIDIA Inception are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.
How it works
1/6
Tell us what you want to check
Describe your dataset and the issues you want to find.
|
training_data.zip0 files
Our plan for your 1 248 files
Find blurry and broken photos
Find emails, phone numbers and IDs in text
Count faces in the photos
Sort into four folders, with reasons
Approve
Fits your request0
Has what you don't want0
For you to check0
Could not check0
1 248 files · 1 248 reasons · 1 sealed record (proof nothing changed)
Fits your request912
Has what you don't want241
For you to check87
Could not check8
Downloads / delivery
Fits your request912 files
For you to check87 files
DatasheetA summary of your files
Sealed recordProves nothing changed
What your AI learns fromDrop the folder here
Illustration · features under development
EU AI Act, Article 10
Good information matters for AI.
Article 10 sets data requirements for providers of high-risk AI systems. The excerpt below concerns training, validation and testing data. Personal data is also subject to the GDPR.
“relevant, sufficiently representative, and to the best extent possible, free of errors and complete in view of the intended purpose.”
Excerpt from Article 10(3) · Read the full text
Errors, flaggedBlur, broken files and copies, flagged with a reason.
Personal details, flaggedEmail addresses, phone and ID numbers in text, and faces counted in photos, before training.
A record for review — plannedThe intended delivery includes a sealed record of the checks and findings: proof that nothing was changed. Customer delivery is not open yet.
Checks can miss issues or flag content incorrectly. They do not establish GDPR or EU AI Act compliance. You decide how to use the results.
What it checks
ImageBlur, copies, exposure and broken files.You see which photos were flagged, and why.
VideoFrozen and dark footage, and a dirty lens, with the time it happens.You see where each problem starts.
Context (text)Email addresses, phone and ID numbers, exposed keys (secret keys and access codes that should not be shared) and copies.You find personal details before training.
Content (things in your photos)The objects you ask for, from our approved list. Faces are counted, never identified.Outlines where available for supported objects in photos.
Questions
What is Dataset Quality Filter?
A dataset cleaning service for AI training data. You send image, video or text datasets and one sentence about what you don’t want. It checks every file, flags quality issues and personal data with a reason, and sorts the files into four folders. It is in development, and we are looking for pilot teams.
Is it a dataset cleaner? Does it delete files?
It does the work of a dataset cleaner without touching your data: it never deletes or changes a file. Each file is flagged with the reason, and you decide what to keep and what to train on.
What does it check in images, video and text?
Images: blur, copies (duplicates), exposure and broken files. Video: frozen and dark footage and a dirty lens, with the time it happens. Text: email addresses, phone and ID numbers, exposed keys (secret keys and access codes that should not be shared) and copies. Content (things in your photos): the objects you ask for, from our approved list; faces are counted, never identified.
Can it find personal data before training?
In text, it flags email addresses, phone numbers, ID numbers and exposed keys (secret keys and access codes that should not be shared). In photos, it counts faces without identifying anyone. Checks can miss issues or flag content incorrectly, so you review the results.
How much does it cost?
Contact us and we will send you a price for your dataset. Purchases open at launch. Contact us
Does it help with the EU AI Act?
Article 10 sets data requirements for providers of high-risk AI systems. The flags, and the record of checks planned for delivery, can support your data governance work. They do not establish GDPR or EU AI Act compliance.
Can I upload a dataset now?
Not yet. Uploads, checks and payments open with the service. Before uploads open, we will present the service terms and a data processing agreement. You can ask about a pilot today. Talk to us about a pilot