SEO Product Content Optimization Platform #
1. Project Overview #
This project aims to build a controlled, auditable, and continuously improving SEO optimization platform for e‑commerce product content.
The initial focus is on WooCommerce-based shopping sites, with future extensibility to all other platforms.
The system is designed to:
- Snapshot and preserve existing product content
- Optimize product titles and descriptions for search engines and target markets
- Score and compare SEO quality before and after optimization
- Enable human review and approval before publishing
- Support recurring (e.g. monthly) re‑optimization cycles based on evolving search criteria
A key design principle is strict separation of concerns:
- Deterministic data handling is performed using Python only
- LLMs are used exclusively for content optimization and evaluation
- Humans retain final authority over publishing decisions
2. Scope (MVP) #
In Scope #
- Shopping products
- English language only
- Product title optimization
- Product long description optimization
- Snapshot-based SEO scoring (before / after)
- Manual approval and publishing
Out of Scope (Initial Phase) #
- Short descriptions
- SEO plugin metadata (Yoast, RankMath, etc.)
- Automated publishing
- Multi-language optimization
- Category and taxonomy optimization
3. High-Level Architecture #
The platform consists of three logical planes:
3.1 Data Plane (Non‑LLM) #
- Fetch product content from shopping platforms via REST API
- Persist snapshots and versions to Supabase
- Publish approved changes back to shopping platforms
3.2 Intelligence Plane (LLM Agents) #
- SEO scoring of existing content
- Content optimization
- Comparative evaluation and improvement analysis
3.3 Control Plane (Human) #
- Review optimized content
- Inspect SEO score deltas and risk flags
- Approve or reject publishing
4. Technology Stack #
Backend #
- Python 3.11+
- FastAPI – API layer and orchestration
- Custom LLM Orchestrator – Agent coordination for optimization and evaluation
(LLM usage follows the LLM Client Interface & Prompt Versioning – Locked Specification (MVP))
Data Layer #
- Supabase (PostgreSQL) – Content snapshots, versioning, scoring, audit logs
Frontend #
- Vue.js – Human review, comparison, and approval interface
External Systems #
- Shopping Platform REST APIs – Source and destination for live product content
5. Optimization Lifecycle (Conceptual) #
- Fetch live product content
- Store baseline snapshot in Supabase
- Score baseline SEO quality
- Optimize content using LLM agents
- Score optimized content using identical scoring criteria
- Compare results and generate improvement report
- Human review and approval
- Publish approved changes
- Repeat process periodically (e.g. monthly)
6. Guiding Principles #
- No direct LLM access to production systems
- All content changes are versioned and reversible
- SEO improvements must be measurable
- Human approval is mandatory for publishing
- Minimize SEO churn by respecting historical performance
7. Future Extensions #
- Multi-language optimization
- Market-specific localization
- Automated keyword trend ingestion
- Partial auto-publishing with confidence thresholds
- Support for additional e‑commerce platforms
8. Related Documents #
- MVP UI Map: 33_mvp_ui_map.md
- Current State & MVP Roadmap: ../backend/docs/41_Current_State_MVP_Roadmap.md
This document serves as a living overview and will evolve as the system design and implementation mature.
