feat: Add reviewiq-pipeline package for LLM-powered review classification
Implement a standalone Python package for processing customer reviews through
a 4-stage pipeline using URT (Universal Review Taxonomy) v5.1:
- Stage 1: Normalization (text cleaning, language detection, deduplication)
- Stage 2: LLM Classification (OpenAI/Anthropic span extraction with URT codes)
- Stage 3: Issue Routing (deterministic issue ID generation, span linking)
- Stage 4: Fact Aggregation (time series metrics for dashboards)
Package includes:
- TypedDict contracts matching Pipeline-Contracts-v1.md
- Async database layer with asyncpg and 5 SQL migrations
- LLM client abstraction supporting both OpenAI and Anthropic
- Sentence-transformers integration for embeddings
- Validation rules V1.x through V4.x
- CLI commands: migrate, run, validate, check
- 55 unit and integration tests (all passing)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
7d720f5378