Platform · Data labeling
A labeling platform your own team can run
Model-assisted pre-labeling, labeling for every data type, review and quality control, and a Data ID Card on every file. Your team does the work, and ours can join whenever you need more hands.
Pre-label
Start from a model's first pass, not a blank file
Run a pre-labeling Twist before anyone opens a file. Foundation models draw the first boxes, masks, tags, and transcripts, and your labelers correct them.
- Pick the model, prompt, labeling type, and output format per run.
- Or host your own model on MLtwist and pre-label at scale.
- Human review of the pre-labels shows where the model is weak.
Label
Label images, video, audio, text, and 3D in one place
Build the ontology once and your team labels every data type against it. Source and labeled video play side by side, so reviewers check tracks and masks as video, not frame by frame.
- Ontology builder with conditional questions that appear only when an earlier answer calls for them.
- Labeler and reviewer roles, per project.
- View 3D scans and meshes with their pre-labels, without converting files.
Review
Review and quality control built in
Approve, reject, or comment on any file, so labelers know exactly what to fix before anything ships. Consensus review and automated checks catch disagreements and anomalies early.
- Thumbs up, thumbs down, or a comment on any file.
- Consensus review when more than one person labels the same file.
- Workload, agreement, and time tracked per person.
Traceability
Every label on the record
Each labeled file carries a Data ID Card: where it came from, the model and prompt that pre-labeled it, who labeled and reviewed it, and every version since.
- Export labels as JSON, with companion files that carry the metadata.
- Deliver back to your storage in the schema your training code reads.
- Each delivery is a version you can diff and roll back.
Integrations
Already use a labeling tool? Keep it.
MLtwist prepares and pre-labels your data, sends it to the tool your team knows, and brings the labels back for review, versioning, and delivery.
Who labels
Your team, ours, or both
The tooling, QA, and Data ID Card are the same whoever does the work.
Your labelers, our platform
Your annotators and reviewers work in MLtwist directly. You set the guidelines and own the queue.
ReadHybridYour experts plus ours
Your specialists take the hard cases and our expert trainers take the volume, in one queue with one QA process.
ReadManagedOur team labels it
Our expert trainers, program managers, and forward-deployed staff run the work on the same platform.
ReadSee your data labeled in MLtwist
Tell us what you're labeling and who's doing it. We'll set up a project on a sample of your data.
Also available through Carahsoft and Google Cloud Marketplace.