AI-powered e-commerce SaaS platform

Metoras

Designed and developed a full-stack SaaS application that helps eBay resellers create, publish, analyze, and improve product listings through AI-assisted workflows and controlled automation.

Visit Metoras

Context: Independent SaaS product
Role: Full-stack product architecture & engineering
Status: Functional private preview
Product: Studio · Copilot
Website: metoras.com (opens in new tab) ↗

01 / Product overview

Two connected workflows for smarter selling.

Metoras brings AI-assisted listing creation and live-listing improvement into one seller workspace.

Studio

Turn product photographs into validated, marketplace-ready listing drafts.

AI-generated listing fieldsImage cleanupeBay validation & publishing

Copilot

Analyze live listings and surface controlled, reviewable improvements.

Listing quality analysisGrouped recommendationsRule-based previews

02 / Engineering architecture

Product interactions stay separate from business logic.

The application uses a workspace-based monorepo with a Next.js frontend, a domain-oriented NestJS API, and a separate worker for asynchronous processing.

PostgreSQL and Supabase services provide the durable data and access layer, while provider integrations remain behind backend boundaries.

Interface layer

Frontend

Browser application deployed on Vercel

Next.js App RouterReactTypeScriptTailwind CSSVercel

Separate runtime processes

Synchronous runtime

Backend API

Domain services and provider boundaries

NestJSDomain-oriented servicesAuthentication and authorizationBusiness logicExternal API orchestration

Asynchronous runtime

Background worker

Long-running work outside browser requests

Asynchronous listing generationCopilot scansAutomation jobsRetry and failure handling

Shared foundation

Data and identity

Durable, owner-scoped application services

SupabasePostgreSQLRow-Level SecurityAuthenticationStorage

Provider boundary

External integrations

Server-side connections behind the API and worker

OpenAIeBay APIsStripeImage cleanup provider

03 / Engineering highlights

Automation built around review and control.

The design treats AI output and external provider responses as inputs to validate, not as authority to change marketplace state.

01Featured control

Human-in-the-loop automation

Copilot groups related issues, supports configurable rules, and presents proposed changes for review before they reach live listings.

AI-assisted listing workflows

Structured listing generation from product images combines content generation, field validation, pricing guidance, and asynchronous processing.

External platform integrations

Backend services coordinate eBay OAuth, account connections, seller policies, listing publishing, and relisting workflows.

Security and data isolation

Encrypted eBay token storage, server-side provider credentials, Row-Level Security, and owner-scoped controls protect seller data and accounts.

Operational reliability

Background workers, job tracking and retries, health monitoring, and cached image-cleanup results address external failures and redundant provider costs.

04 / Technical challenges

Design decisions at system boundaries.

Marketplace constraintsCoordinate probabilistic AI-generated content with eBay’s strict category, field, policy, and publishing requirements.
Reviewable automationKeep potentially risky changes visible and reversible through grouped recommendations and preview-first workflows.
Async failure handlingTrack long-running jobs, retries, and external API failures without tying processing to a browser request.
Connected-account securityProtect seller data, OAuth tokens, and provider credentials while enforcing owner-scoped access.
Clear application boundariesSeparate frontend interactions from backend business rules, provider orchestration, and durable data access.