# amazon-po-parser

Tools for ingesting, scraping, and enriching Amazon purchase order data from Coupa emails and Payee Central.

## Architecture

- **Data source**: Coupa PO notification emails → parsed into structured JSON
- **Enrichment**: Playwright scrapers hit Amazon Payee Central to resolve site codes, ship-to addresses, and line items
- **Storage**: DynamoDB `purchase-orders` table (us-east-1) is the single source of truth
- **Mapping**: `output/site-state-extra-mapping.json` — curated Amazon facility code → US state mapping (392 codes)

## Scripts

| Script | Purpose |
|---|---|
| `parse-mbox.py` | Parse Coupa PO emails from MBOX export into structured JSON |
| `scrape.mjs` | Single-browser Payee Central scraper |
| `scrape-payee-chunk.mjs` | Chunk-based concurrent scraper (run N instances in parallel) |
| `scrape-fill-chunk.mjs` | Lightweight scraper for backfilling site codes and ship-to addresses |
| `gen-po-list.mjs` | Generate PO lists from invoice spreadsheets |
| `monitor.mjs` | Monitor scraper progress across chunks |

## DynamoDB Schema

Table: `purchase-orders` | Key: `po_number` (string)

| Field | Type | Source |
|---|---|---|
| `po_number` | string | Email/Coupa |
| `site_code` | string | Payee Central scrape |
| `state` | string | Derived from site_code mapping |
| `ship_to_raw` | string | Payee Central scrape |
| `line_items` | list | Email parse + Payee Central |
| `trade` | string | Classified from descriptions |
| `trades` | list | All trades for multi-line POs |
| `fiscal_year` | string | Order date year |
| `coupa_category` | string | Coupa commodity category |
| `source_system` | string | "coupa" |
| `data_source` | string | "email+payee_scrape" |
| `invoice_total` | string | Downstream invoice matching |
| `invoice_count` | number | Downstream invoice matching |
| `order_date` | string | Email parse |
| `po_status` | string | Email parse |
| `payment_terms` | string | Email parse |
| `requisition_number` | string | Email parse |

## Prompt

`prompts/po-email-parser.md` contains the LLM prompt for parsing incoming Coupa PO emails. It includes site code extraction patterns, a skip list for non-site codes, trade classification rules, and ship-to address parsing guidance.

## Setup

```bash
npm install          # playwright
```

Requires a Chrome profile logged into Amazon Payee Central at `./chrome-profile/`.