Andrew Parra ES

Lab · Shopify

Trendline

A Shopify app: find a product worth testing, then have Claude build a branded one-product store around it the same afternoon.

Status
Live
Who uses it
Andrew’s own shop, through a private copy of the app. The public version is built to stay eligible for the Shopify App Store.
Built with
TypeScript on Netlify Functions + Blobs · Shopify Admin GraphQL, webhooks, Billing API · Claude with web search

Four scans merge into one list

Example

Trend risers

Sunset lampWalking pad

TikTok / Meta virality

Sunset lampPet hair roller

Amazon rank jumps

Portable blenderWalking pad

Reddit “I wish this existed”

Cable organizer box

↓ merged · duplicates collapsed · each scored

  1. Sunset lamp2 lanes → 182
  2. Walking pad2 lanes → 177
  3. Pet hair roller71
  4. Portable blender64
  5. Cable organizer box58

Autopilot progress log

  1. ✓ scan
  2. ✓ supplier check
  3. ✓ pick
  4. ✓ write
  5. ✓ build
Draft: nothing published
Four research passes run in parallel with live web search and merge into one de-duplicated list, scored on demand, margin, shipping-friendliness and seasonality.

The problem

Launching a test store means research, sourcing, copywriting, policies, theme work and ad copy, which usually takes days. Trendline compresses it into a scan and one launch, with merchant-safe defaults.

How it works

Scan (this was “Product Scout”)

Four research passes run in parallel with live web search: trend risers, TikTok and Meta virality, Amazon rank jumps and Reddit “I wish this existed” threads. They merge into one de-duplicated list scored on demand, margin, shipping-friendliness and seasonality.

Sourcing

Real supplier listings and landed cost, with Claude double-checking the listing is the same product and not a part or an accessory.

Launch

Claude writes a brand, the product page, store pages, policies and ad copy; Trendline creates them in the store on a forked Dawn theme. Draft launches publish nothing.

Autopilot

Scan → supplier check → pick → write → build a draft store as one job, with a readable progress log.

Fulfilment stays manual

Nothing is ordered automatically. The merchant sees the cost and approves each supplier order on their own supplier account.

Daily store review

A scheduled review of Andrew’s own store proposes changes and lessons; only Andrew approves them.

Where AI fits

  • Claude with web search does the research, brand, copy and supplier matching.
  • Every request is built and metered on the server: the browser names a task and can never choose the model, the prompt or the tools.

Guardrails

Spend limits first

Plan limits are charged before any AI tokens are spent, with a hard monthly AI-cost ceiling per shop.

No invented social proof

Generated copy may never invent reviews, ratings, customer counts, certifications, discounts or “was” prices.

Hands off what exists

A merchant’s existing policies and menus are never overwritten without a tick-box.

Minimal customer data

Names, addresses and phones are never stored, order records are deleted after 90 days, and uninstalling wipes everything.

Engineering notes

Closed an open proxy

AI calls moved server-side after a client-built request was found to bypass plan limits.

Merchant stays buyer of record

Supplier orders always run on the merchant’s own key, never Trendline’s.

Two apps, one backend

A private copy runs on Andrew’s shop so changes are proven there before the public app gets them.