Description
Audience: IR engineers, knowledge managers, RAG builders.
What you get: taxonomy starters (tech/finance/healthcare/policy) as extendable YAML; few‑shot/zero‑shot templates with acceptance thresholds; confidence accounting (scores, abstain bucket); exporters (CSV/Parquet/JSONL) with lineage stamps; QC toolkit (spot‑check sampler, drift detector, confusion matrix).
Architecture: spaCy/HF pipelines, optional embedding‑based label expansion, OpenSearch ingest mappings.
KPIs: tag coverage, label accuracy, retrieval uplift (nDCG@K, MRR). Guardrails: bias review, override pathways, versioned taxonomies with rollback.

Reviews
There are no reviews yet.