Data for every stage of model training

Data collection

Speech, text, image and video gathered to your spec, with consent and provenance recorded for every sample.

• Scripted and spontaneous speech in 60+ languages and accents
• Image and video capture with diverse subjects, devices and settings
• Domain text written by subject-matter experts
• Signed consent and metadata recorded per sample

Common uses: Speech, text, vision

Annotation and labeling

Bounding boxes, segmentation, transcription, entity tagging and more, reviewed in two passes before delivery.

• 2D and 3D bounding boxes, polygons and semantic segmentation
• Audio transcription, speaker diarization and timestamping
• Named entities, intent, sentiment and relation labeling
• Two-pass review with measured inter-annotator agreement

Common uses: Vision, NLP, audio

RLHF and alignment

Preference ranking, rubric scoring and expert-written demonstrations that teach models how to respond.

• Pairwise and multi-response preference ranking
• Supervised fine-tuning demonstrations written by experts
• Custom rubrics for helpfulness, accuracy and tone
• Reviewers with STEM, legal, medical and coding backgrounds

Common uses: LLMs, agents

Model evaluation

Side-by-side comparisons, red-teaming and safety reviews that show where a model fails before users find out.

• Blind side-by-side model comparisons
• Adversarial red-teaming for safety and policy gaps
• Factuality and hallucination audits
• Benchmark datasets built for your use case

Common uses: Safety, quality

Agent and coding data

Multi-step task trajectories, tool-use traces and reviewed code solutions written by vetted engineers.

• Multi-step task trajectories in browsers and apps
• Tool-call and function-use verification
• Code solutions, reviews and unit tests from vetted engineers
• Step-level grading of agent reasoning

Common uses: Agents, code

Physical AI data

Sensor fusion, LiDAR point clouds and robot demonstration labeling for autonomy and robotics teams.

• LiDAR and radar point cloud labeling
• Multi-camera sensor fusion and object tracking
• Robot manipulation demonstration annotation
• Edge-case scenario mining for autonomy

Common uses: LiDAR, robotics