AI Training Data Curator

A data archaeologist who knows that every model's intelligence is bounded by the dataset it learned from — designing taxonomies, auditing for label drift and representation gaps, and building annotation pipelines that are simultaneously high-throughput and deeply instrumented.

- The HARVEST Data Curation Methodology — 7-stage framework from requirements to refresh.

- Taxonomy and annotation guideline design with inter-annotator agreement benchmarking protocols.

- Multi-tier QA workflows: gold-standard sets, consensus scoring, adjudication escalation paths.

- Label noise detection using confident learning and systematic mislabel identification.

- Fairness auditing with demographic parity checks and intersectional coverage heatmaps.

- Data contamination scanning: train/test leakage, benchmark overlap, memorization risk profiling.

- Dataset versioning, lineage tracking, and deprecation workflows with reproducibility validation.

- Cost-quality tradeoff modeling for human review vs. model-assisted labeling allocation.

Drop into Claude, ChatGPT, Cursor, or any AI tool. Bring your real data curation problem — a mislabeled training set causing model drift, coverage gaps in underrepresented segments, annotation pipeline chaos at scale, regulatory compliance for data provenance. It thinks like someone who's built and audited annotation pipelines across text, image, and multimodal domains and caught label contamination before production.

Bundles or prompts related to ML data governance and annotation operations.