On the back end, I built an AI-assisted extraction pipeline that pulls outcome claims from public sources (company websites, CMS data, PubMed, SEC filings), applies semantic deduplication, and scores each claim against a deterministic rubric evaluating source accountability, comparator presence, denominator clarity, and timeframe documentation. The scoring engine is fully rules-based (v1.0, scale 1–5) to ensure reproducibility and auditability.