Services & delivery
What we do, and exactly what you receive.
Purchase any service on its own or combine them into a pipeline. Every engagement resolves into the same structured delivery package, checked against defined acceptance criteria.
Core path
Data Collection
Task/scene design, recruiting, equipment deployment, teleoperation and on-site work, trial capture, batch capture and recapture.
You receive
Raw data, task/device metadata, collection logs, batch lists and quality reports.
Annotation & Processing
Detection / segmentation / tracking, 2D/3D pose, action segmentation, language/speech, expert labels and multimodal associations.
You receive
Label files, annotation guidelines, version history, review results and hard-case lists.
Data Engineering
Cleaning / deduplication, time synchronization, coordinate/unit conversion, invalid-frame handling, dataset splitting and format adaptation.
You receive
Versioned datasets, data dictionaries, reading examples, split lists and conversion notes.
Support & operations
Experts & Evaluation
Expert task design, answer/trajectory evaluation, preference ranking, reward/failure labeling and independent held-out sets.
You receive · Expert feedback, scoring rules, evaluation sets, item-level results, disagreement arbitration and analysis.
Quality Management
Sampling inspection, gold examples, duplicate annotation, independent review, issue tracking and rework management.
You receive · Inspection reports, agreed error/agreement measures, rework records and batch acceptance sheets.
Simulation & Synthesis
Scene/object assets, parameter randomization, rendering, trajectory generation, auto-labels and human review.
You receive · Scene assets, synthetic data, generation configuration, source identification and version and quality notes.
Managed Teams
Project-configured collection operators, teleoperators, annotators, QA, project managers and domain experts.
You receive · Ongoing output, staff training, scheduling, progress and quality reports and task handoff.
Delivery package
Five layers, each with its own acceptance check.
Data & labels
Raw files, processed data, labels and sample previews.
Acceptance · Readable files, complete counts and one-to-one correspondence between labels and samples.
Fields, space & time
Field dictionary, units, coordinate frames, timestamps, calibration and synchronization explanation.
Acceptance · Temporal consistency, coordinate conventions, calibration versions and measured error.
Provenance & versions
Task/scene/device/anonymized-person IDs, batches, source and change records.
Acceptance · Traceability, deduplication and distinction between original and derived data.
Quality & splits
Label specification, review records and train/validation/test lists.
Acceptance · Holdouts by person, scene, object or task to control data leakage.
Usage & reading
License scope, data description, format version and example loading code.
Acceptance · Agreed use and scope, successful read and full understanding of fields.
Where to next