MULTI-TENANT SAAS / PRIVATE SOURCE
Fandem AI.
Arabic-aware AI support for MENA e-commerce.
The project
A multi-tenant SaaS platform with an embeddable AI chat widget for MENA e-commerce. A Vanilla TypeScript Web Component uses Shadow DOM to integrate into storefronts, backed by NestJS services and a Next.js dashboard. Arabic-aware retrieval combines Cohere embeddings, PostgreSQL and pgvector, with intent routing across Gemini models through the Vercel AI SDK.
My contribution
Built the embedded widget, backend, dashboard, and retrieval workflow.
How I approach it
- Use Shadow DOM to keep the embedded widget isolated from a storefront’s styles.
- Ground Arabic-aware responses in retrieved context using Cohere embeddings and pgvector.
- Share Zod schemas and Drizzle types across a monorepo, with Docker for reproducible development.
NestJSNext.jsPostgreSQLpgvectorDrizzleTypeScript