B2B Multi-Vendor Marketplace

Problem

A B2B sourcing platform needed an end-to-end marketplace where verified buyers could solicit bids from verified sellers via reverse auction, transact with built-in KYC and dispute resolution, and accept AI-assisted product listings. The existing off-the-shelf marketplace plugins didn't cover reverse auctions, complex seller verification flows, or AWS-Lambda-backed product enrichment.

Solution

Built on top of WordPress + WooCommerce + Dokan, extended with 17+ custom plugins that own the marketplace's distinctive flows: reverse-auction bidding, multi-step seller KYC, AI product creation (an AWS Lambda that takes a few keywords and produces SEO-friendly descriptions), live chat between buyer and seller, and a dispute workflow with admin escalation. Infrastructure runs in Docker Compose with Apache + LiteSpeed in front, MariaDB primary + Redis object cache, and a GitLab CI/CD pipeline that does backup → deploy → rollback automatically on every push to main.

Architecture

• WordPress as the CMS + auth layer; WooCommerce for catalog/cart/checkout; Dokan for multi-vendor.
• Custom plugins layered on Dokan's hooks for reverse auctions, KYC stages, dispute states.
• AWS Lambda function fronted by API Gateway for AI product enrichment (called from the seller-side admin UI).
• Redis as a session and object cache; MariaDB as primary DB with a nightly logical backup to S3.
• GitLab CI: lint → build Docker images → push to registry → SSH deploy with health-check rollback.

Key technical decisions

• Custom plugins over forks — kept Dokan's core upgradeable; all bespoke logic lives in side-by-side plugins that hook into Dokan rather than patching it.
• Docker Compose over a managed PaaS — gave full control over LiteSpeed cache config and the WordPress object cache, both of which matter at scale on this stack.
• Lambda for AI — enrichment can be slow and fail; isolating it as a stateless function meant the WordPress request never blocked on it.

Outcomes

• 17+ custom plugins shipped, fully deployable via CI/CD with automated rollback.
• AI product creation cut seller listing time noticeably; KYC + dispute flows handled in-house with no third-party SaaS.

What's next

• Migrate the AI enrichment Lambda from prompt templates to a managed Bedrock workflow.
• Move primary cache from Redis to a tiered LiteSpeed + Redis combination for content vs object cache separation.