Prashanthvara Catalog Enrichment Pipeline
Takes a raw supplier catalog feed and produces an enriched, normalized, scored catalog: consistent descriptions, mapped categories, flagged gaps, and a supplier-facing remediation report.
Inputs
- Supplier feed: CSV/TSV with whatever columns the supplier sent
- Target taxonomy: the category tree items must map into
- Enrichment sources allowed: supplier API, manufacturer part number lookup, structured extraction from description text
Synthetic sparse feed in references/sample_feed.csv.
Workflow
- Run
scripts/profile_feed.pyto produce a completeness matrix: per field, fill rate and format consistency. - Normalize units of measure, pack sizes, and manufacturer part numbers using the rules in
references/normalization_rules.md. UOM ambiguity (EA vs CS vs PK) is the highest-frequency ordering-error source; resolve it explicitly. - Map items to the target taxonomy. High-confidence mappings apply automatically; low-confidence mappings go to a review file with the top-3 candidate categories and the evidence for each.
- Score the enriched catalog and emit the supplier remediation report: fields only the supplier can fix, ranked by order-impact.
Output format
Enriched catalog CSV, ## Completeness Report (before/after matrix), review_queue.csv, supplier remediation report (markdown, sendable as-is).
Guardrails
- Never fabricate product attributes. Enrichment fills from a source or flags the gap.
- Auto-applied category mappings require a stated confidence threshold, logged per item.