Real automations, real numbers.

Challenge

Six retail locations were each running independent reorder workflows in spreadsheets. The owner spent every Sunday compiling supplier emails and chasing missed deliveries. Errors compounded: wrong quantities, late orders, weekend rush fees.

Approach

We built a Python + Django reorder engine connected to their POS, with rules per supplier and per location. Low-stock signals trigger draft POs; suppliers receive them via email or portal scrape (Selenium); confirmations come back as PDFs that we OCR and write to Postgres. Slack notifies the manager only when something needs human judgment.

Stack

Challenge

The store was bleeding Google Shopping placements daily because Merchant Center kept disapproving listings for policy + missing-attribute reasons. Manual cleanup couldn't keep up. Every new variant created a new disapproval.

Approach

We built a continuous feed pipeline: GMC API + Shopify Admin GraphQL, with Claude rewriting non-compliant titles and descriptions, and GPT-4 Vision regenerating images that failed image-quality policy. A nightly job diffs the feed and only pushes changes. Top 50 SKUs by revenue go through a human approval queue first.

Stack

Challenge

The bookkeeper spent 4–6 hours every weekend reconciling cash skim, tips, and payment processor statements against the POS. Missing receipts and miscategorized transactions cascaded into month-end pain.

Approach

We built an ETL pipeline that pulls POS data nightly, parses payment statements with OCR + GPT-based metadata validation, and reconciles automatically. Mismatches go to a small review queue. Power BI dashboards give the owner a live operational picture; the bookkeeper gets a clean export for QuickBooks.

Stack