Every engagement ships these as concrete artifacts you own — not slides, not hand-waving.
Items in your schema (COCO, YOLO, JSONL, custom) delivered to S3 / GCS / your endpoint. Versioned, manifest-tracked, deduplicated.
We run your existing model (or a stock SAM 2 / Florence-2) on raw items first; humans correct rather than label from scratch. Throughput up 3–5×.
Double-blind sampling at 5–15%, error taxonomy, weekly drift reports. We ship the audit trail along with the data.
Versioned annotation guidelines with edge-case galleries. New annotators onboard from this in <2 days.
Lock the label taxonomy, edge cases, and acceptance criteria. Run a 100-item pilot batch. Adjust guidelines from real annotator feedback.
Onboard the pod, run calibration batches until inter-annotator agreement crosses your threshold (typically ≥95%).
Daily / weekly batches at agreed rate. Live dashboards (volume, agreement, error rates, throughput). Weekly QA reports.
Monthly distribution reports, flagged outliers, schema-revision recommendations as your data evolves.
Best-in-class where it matters; boring and battle-tested everywhere else.
Pricing depends on modality, schema complexity, and QA stringency. Bounding-box image at the floor; instance-segmented video at the ceiling. Volume tiers above 1M items / month.
Image (boxes, polygons, segmentation, keypoints), video (tracking, action), LiDAR (3D boxes), text (NER, classification, RLHF preference), and audio (transcription, speaker diarisation, sentiment).
Yes — DPA / NDA standard, work happens on locked-down workstations, no data leaves the office network. We're ISO 9001 certified and operating under several BFSI-grade scopes.
No. Your data is yours, contractually never re-used or aggregated with other clients' data.
They get escalated to a senior annotator and added to the guideline gallery. Recurring ambiguities trigger a schema-revision call with you.