AZParts — Web + WhatsApp Auto Parts Marketplace

AZParts is a production marketplace for used auto parts in Saudi Arabia. Instead of a buyer calling scrapyards one by one, the platform broadcasts the request and lets specialized scrapyards compete on price. The customer journey runs as two complete flows: a full web marketplace, and a WhatsApp flow where a stateful bot handles search, premium sourcing requests, payment links, order status, and AI-assisted support. AZParts hired me to build it, then hired me again to run it — I’m still the one keeping it running today.

Finding a used part in Saudi means calling dozens of scrapyards, negotiating blind, and hoping the part shows up. Dealers, meanwhile, have inventory but no digital channel their customers actually use.

A bidding marketplace with two complete flows — a full web experience, and the channel everyone already uses, WhatsApp. Buyers describe their car and part once; the system routes the request to matching scrapyards, collects competing offers, takes payment (one-time or installments), and manages delivery — with an AI support layer answering questions from a knowledge base.

• Scrapyards bid on each request, so buyers get the lowest real price instead of negotiating alone.
• Two complete journeys — search, offers, checkout, delivery updates — one on the web, one entirely inside WhatsApp.
• Premium paid-search requests, refunds, and delivery assignment run on auditable, queue-backed workflows.

• Own the platform end to end as product engineer: web app, WhatsApp flows, payments, and deployment.
• Built resilient webhook processing with signature validation, idempotency, and dead-letter capture for every provider.
• Designed the asynchronous inbound queue (Redis + BullMQ workers) that keeps WhatsApp conversations fast under load.
• Integrated EdfaPay (one-time) and Tamara (installments) with webhook reconciliation and refund workflows.
• Shipped the retrieval-augmented AI support flow (OpenAI + pgvector) with runtime-configurable limits.

• Guided vehicle search (make / model / year) with competing offers from scrapyards
• Stateful WhatsApp bot: part search, premium requests, checkout links, order tracking, human escalation
• Premium paid-search lifecycle: intake, service fee, status updates, refunds, notifications
• Delivery request APIs with driver assignment workflow
• AI support chat grounded in platform knowledge (RAG)
• Audit logging with export, and health checks across Redis/PostgreSQL

• Keeping conversational state coherent across a channel users treat as free-form chat.
• Reconciling payment provider webhooks reliably enough to run refunds without manual cleanup.
• Serving an Arabic-first audience with flows that still work for non-Arabic speakers.

Next.js (App Router) · Node.js · PostgreSQL + pgvector · Prisma · Redis + BullMQ · Twilio WhatsApp API · EdfaPay · Tamara payments · OpenAI (RAG support) · Docker + Caddy