Enrichment Agent Dashboard
Sign in with your API key to continue
Your API key is stored in sessionStorage only — cleared when you close the tab.
| Product | SKU | Scenario | Confidence | Status | Created | |
|---|---|---|---|---|---|---|
|
Loading jobs… Fetching enrichment queue from server |
||||||
Live pipeline events — ring buffer of last 500 entries.
Logs auto-refresh every 15s. Showing up to 200 entries. Reset when the server restarts.
Multi-agent LangGraph workflow — each job passes through all 5 stages in sequence.
Retry loop: If QA confidence is below 60%, the pipeline loops back to Data Extraction for a second pass (max 1 retry via LangGraph conditional edge). Jobs with confidence ≥85% and no TGA flags go straight to review queue; 60–85% jobs go to review with flagged fields highlighted.
Own-brand products with a PDF or DOCX tech sheet attached. Parsed with PyMuPDF / python-docx, then passed to Claude Sonnet for structured data extraction.
Supplier / third-party brands. Anthropic's built-in web search finds the product page; Claude extracts structured ingredient and dosage data.
No tech sheet and no researchable brand site. Claude Vision OCRs the product label images concurrently. Most likely to need manual review and editing.
Preferences are saved in your browser (localStorage).
Register a products/update webhook so Shopify automatically queues enrichment jobs when products are tagged. The address must be publicly reachable (use ngrok or your production URL).
Loading…
Agent model and pipeline parameters are set via environment variables — edit .env and restart the server to change them.