dataset-redaction

What problem does it solve?

This Skill creates safe evaluation datasets by redacting PHI and optionally generating synthetic equivalents while preserving document structure necessary for OCR, STT, and LLM benchmarking.

Core Features & Use Cases

• Redaction: apply deterministic pseudonymization to patient identifiers across visits.
• Synthetic generation: produce realistic, test-ready data with controlled deltas for benchmarking.
• Deliverables: provide ready-to-use artifacts such as redaction policies, schemas, and tooling specs for reproducibility.

Quick Start

Run the redaction pipeline on a sample dataset to generate redacted_documents.json and gold_facts.json, then validate the dataset with the provided manifests.