Project overview

About the project.

Our client set out to solve two problems that have quietly limited the Saudi automotive parts market for years: buyers rarely know the exact part they need, and suppliers rarely hold the exact quantity anyone wants. We built a platform that addresses both. At the front, an AI-powered Search Coordinator identifies parts from whatever the buyer actually has — a description in plain language, a photograph of a worn component, a voice note, a scanned invoice, or a VIN — and returns ranked candidates with confidence scores, asking targeted clarification questions instead of guessing when the input is incomplete. Behind it sits a procurement engine built for volume: customer requests are split into line-level demand, aggregated into product-level demand buckets across all buyers, frozen into supplier RFQ batches, sourced from single or multiple suppliers with automatically blended landed costing, and allocated back to each original customer as one clean quote package. The buyer sees a simple price, lead time and validity date. The supplier sees only their own request. The admin sees the entire chain, fully audited, from customer request to supplier response to final margin. The result is a platform where a team can process hundreds of concurrent quote requests with the same headcount that previously handled dozens.

AI-integrated vehicle parts search and multi-supplier quoting platform — Kingdom of Saudi Arabia Richinnovations
Technical specifications

Under the hood.

BACKEND & ARCHITECTURE NestJS (Node.js / TypeScript) built as a modular monolith on Domain-Driven Design principles. Every domain — user, supplier, admin, orders, payments, inventory — encapsulates its own controllers, services, entities and interfaces over a shared database, giving strict separation today and clean microservice extraction tomorrow without refactoring.
AI INTELLIGENCE LAYER An independent Search Coordinator service in its own repository, handling multi-modal part identification across text, image, audio, OCR and VIN input. Returns ranked candidates with confidence scores through a stable structured JSON contract, with configurable business thresholds, session-memory refinement loops and full retry and timeout handling.
PERFORMANCE & SCALE Redis caching and queue-based asynchronous processing for supplier notifications, quote generation and long-running workflows, backed by backend-enforced role permissions, immutable audit logging, and load and integration testing prior to production release.
More case studies