Commercy.ai

AI-Powered Tools for Shopify Success

Visit the website

90%
faster product description creation with Meta Magic
AI-powered
Visual and semantic product search powered by AI
10+
languages supported for generated product descriptions

Client

Commercy.ai

Visit the website

Industry

E-commerce Technology

Client overview

Commercy.ai is a startup building AI tools for Shopify merchants. The idea: automate the routine store work so merchants can spend their time on the parts a human has to do.

Problem statement

Shopify merchants lose hours to two chores: making products findable and writing descriptions for them.

Context

A store with bad search loses sales it never sees, and a store with thin descriptions ranks worse and converts worse. Both problems grow with catalog size.

Pain points

Poor product search experience limiting conversions
Manual creation of product descriptions consuming time and resources
Difficulty maintaining consistency and SEO across product catalogs

Approach

Build a small suite of Shopify plugins that take over the search and content work, each installable on its own

Technologies used

Artificial Intelligence and Machine Learning, Natural Language Processing, Image and Text-Based Semantic Search, Multilingual Generation, Shopify API, Shopify Admin API, Shopify Polaris Components, Next.js, PostgreSQL, Vector Database

Implementation details
Phase 1. Initial Assessment
- Identified key areas where AI could significantly impact merchant operations, focusing on search functionality and content creation.
Implementation details
Phase 2: Development/ Implementation
- Searchy AI: An advanced AI search engine that combines semantic text understanding and natural language processing. Customers can type queries to find the most relevant product match. It understands context, synonyms, and even typos to return accurate results that boost discovery and conversion. Implemented as a Shopify plugin using Shopify APIs, Polaris components, and Next.js with a PostgreSQL and vector DB backend.
- Meta Magic: A content automation engine that creates engaging, SEO-optimized product descriptions and meta tags. Merchants choose tone of voice, target market language, and more. Outputs are high-converting, multilingual, and tailored to product specifics. Built as a Shopify plugin leveraging Shopify Admin API and integrated with the merchant store backend.
Implementation details
Phase 3: Testing and Launch
- Tested against real merchant catalogs, then launched with onboarding guides built into the plugins.

Key outcomes

+Conversion

+Conversion driven by improved search and discovery

90%

90% time savings on content creation tasks

Global reach

Global reach enabled through multilingual content support

Client feedback

Marcus Sepp

A Shopify merchant

Customers actually find what they type now, typos and all. And Meta Magic writes in an evening what used to take us a week

Tech Stack

Backend
- Supabase Postgres + Auth for primary data and identity
- Inngest for jobs, queues, scheduling; Redis for caching, pub/sub, and rate limits
- Self-hosted LLM inference (secure, isolated) and self-hosted vector DB for embeddings + semantic search
- Shopify Admin/Storefront GraphQL + Webhooks for sync and app flows; Postmark for transactional email
Frontend
- Next.js (App Router, Server Components, API routes)
- React UI following Shopify Polaris patterns
- Embedded app via Shopify App Bridge with OAuth/session tokens
- Internationalization and rich editor workflows for product content
Infrastructure
- Containerized services for LLM, vector DB, and Redis
- Supabase managed DB, storage, and auth
- Environment-scoped config and secrets management
- Webhook processing with retries and DLQs orchestrated by Inngest
Security
- Data isolation: all LLM/vector processing remains on self-hosted services
- Supabase Row-Level Security and strict API route guards
- Token-scoped Shopify API access, key rotation, and HMAC webhook verification
- Redis-backed rate limiting plus audit logging for abuse prevention

Key takeaways

Lessons learned

Generic AI tools don't survive contact with a real store. These worked because they were built around specific merchant chores: descriptions, search, translations. Nothing broader.