Available for new work

Abdulrahman Faraj — Amman, Jordan

Building AI agents thatsell, book, answer, operate

AI engineer and full-stack developer with four agent products live in production — RAG over private data, tool calling into real systems, WhatsApp, web, and voice channels, and the multi-tenant plumbing that keeps it all running unattended.

View workGitHub(opens in a new tab)
04
AI agents in production
11
Systems built end to end
24/7
Running unattended
AR
Arabic-first by default
Next.jsTypeScriptPythonFastAPINestJS.NETClaude APIGeminiOpenAIVercel AI SDKLangChainMCPRAGElevenLabsPostgreSQLpgvectorPrismaDrizzleRedisBullMQWhatsApp Cloud APISocket.IOOpenTelemetryDockerGitHub Actionsn8nPlaywrightFlutter

Selected work

Four products, live in production.

Arabic-first AI across enterprise chat, marketing, WhatsApp booking, and private assistance — each one built end to end and running for real businesses.

Also built

More systems, same discipline.

Commerce agents, NL-to-SQL answering, multi-agent pipelines, and the full-stack work that sharpened the craft — some of it on its way to production.

  • Kammelكمّل

    WhatsApp revenue automation

    Arabic-first WhatsApp revenue automation for Saudi small businesses — engineered like a production system from day one, with job queues, tracing, and metrics built in.

    Next.jsDrizzlepg-bossOpenTelemetry

  • Tammتمّ

    WhatsApp commerce agent

    An Arabic AI sales agent that sells a merchant's catalog inside WhatsApp — browsing, cart, delivery, and “where is my order?” — on a multi-tenant commerce engine.

    NestJSPrismaWhatsApp Cloud APIMulti-tenant

  • Shamsiشمسي

    Database-grounded answering agent

    Answers questions about education in Jordan from the database alone: Gemini turns the question into guarded SQL, rows produce the facts, and code verifies before anything is said.

    GeminiNL → SQLWhatsAppNode.js

  • Luma Architectلوما

    Multi-agent blueprint platform

    Turns a raw product idea into a complete software blueprint — analysis, architecture, and documentation produced by cooperating agents in one pipeline.

    Multi-agentNode.jsPrismaPrompt pipelines

  • Obscura Studioأوبسكيورا

    Agentic video production

    An agentic Arabic YouTube pipeline: one approved topic becomes a researched long-form video — script, licensed assets, ElevenLabs narration, and a data-driven Remotion edit.

    Claude CodeAgentsElevenLabsRemotion

  • SkillBridgeسكيل بريدج

    Talent platform, clean architecture

    A hiring platform with separate job-seeker, company, and admin experiences — an ASP.NET Core clean-architecture API with a tested domain core and a React client.

    ASP.NET CoreC#ReactVite

  • Anime Zoneأنمي زون

    Streaming library & player

    A cinematic anime library with a self-populating AniList catalog, an HLS player with resume and multi-server switching, and a full admin CMS.

    Next.jsPostgreSQLDrizzleVidstack

What I build

Systems that hold up after launch day.

Four kinds of work, all of them shaped by the same constraint: it has to run in Arabic, for real customers, without me in the loop.

  • AI agents & assistants

    Agents that call real tools against real systems — with the guardrails, retries, and observability that keep them from embarrassing you in front of a customer.

    • Tool loops & function calling
    • MCP integrations
    • Fallbacks and escalation to a human

  • Arabic-first NLP & retrieval

    Retrieval that works on Arabic the way it is actually written — dialects, diacritics, mixed scripts — instead of pretending it is English with different glyphs.

    • Chunking & embeddings for Arabic
    • pgvector search
    • Evaluation on real transcripts

  • WhatsApp & channel automation

    The Gulf runs on WhatsApp. I build the booking, reminder, and support flows that live inside it — template approval, webhooks, and delivery states included.

    • WhatsApp Cloud API
    • Scheduling & reminders
    • n8n orchestration

  • Multi-tenant SaaS engineering

    The unglamorous half: tenant isolation, migrations, rate limits, deploys that don't wake anyone up. This is where most AI demos quietly fail.

    • Row-level tenant isolation
    • Next.js + FastAPI services
    • Docker & GitHub Actions

How I work

Narrow the problem, then ship it end to end.

Same four steps on every project, whether it's a two-week automation or a platform with tenants.

  1. 01

    Frame

    Find the one workflow worth automating and write down what "working" means in numbers. Most requests shrink by half at this step — that's the point.

  2. 02

    Architect

    Pick the smallest architecture that survives growth: where tenants split, what the model is allowed to touch, and what happens when it's wrong.

  3. 03

    Ship

    A narrow version reaches real users early. Arabic is tested from day one, not localized at the end — that ordering is the difference.

  4. 04

    Operate

    Watch real transcripts, fix what actually breaks, and keep the deploy path boring. A system nobody has to babysit is the deliverable.

Stack

Chosen for deploy speed and tenant safety.

Nothing here is picked for novelty. Each tool earns its place by shortening the path from a change to production — and everything listed has shipped in a real system.

Languages

Typed where it counts

  • TypeScript
  • Python
  • JavaScript
  • C#
  • Dart
  • SQL

Frameworks

Server-first by default

  • Next.js
  • React
  • FastAPI
  • NestJS
  • Express
  • ASP.NET Core
  • Node.js
  • Flutter

AI / Agents

Model-agnostic by design

  • Claude API
  • Gemini
  • OpenAI
  • Vercel AI SDK
  • LangChain
  • MCP
  • RAG
  • ElevenLabs

Data

Postgres unless proven otherwise

  • PostgreSQL
  • pgvector
  • Drizzle
  • Prisma
  • Supabase
  • Redis
  • Firebase

Infrastructure

Boring on purpose

  • Docker
  • GitHub Actions
  • Vercel
  • Nginx
  • PM2
  • n8n
  • BullMQ
  • OpenTelemetry
  • Playwright

Channels

Where MENA users already are

  • WhatsApp Cloud API
  • Web chat
  • Voice
  • Telegram · Discord
  • Email · Calendar

About

From Amman, for the Arabic-speaking market.

I'm Abdulrahman Faraj — a full-stack developer and AI engineer. I've built eleven systems end to end, and four Arabic-first AI agent products are live in production today.

My work sits at the boring, load-bearing end of AI: tenant isolation, WhatsApp automation, retrieval that actually handles Arabic, agent tool loops, and deploy discipline. A demo takes a weekend; a system that runs unattended for a year is a different craft.

The interface stays quiet on purpose. The proof is in what's shipped.

Based in
Amman, Jordan
Focus
AI agents, built end to end
Market
MENA & the Gulf — Arabic-first AI SaaS
Working languages
Arabic (native) · English (professional)

Journey

  1. 2026

    Safa — local-first assistant

    Shipped a private assistant that runs entirely on the user's machine and connects to their inbox, calendar, and drive over MCP.

  2. 2025

    Mawid — WhatsApp booking

    Took a dialect-aware booking agent to production for Gulf clinics and salons, on top of the WhatsApp Cloud API.

  3. 2024

    Botify & Campify

    Built two multi-tenant platforms — enterprise chat agents and AI marketing campaigns — with retrieval and tool calling at the core.

  4. 2023

    Going product-first

    Committed to building Arabic-first software as products rather than one-off client work — the discipline the four live systems came from.

Contact

Build quietly. Ship clearly.

For Arabic AI automation, MENA SaaS, or full-stack product engineering — write directly. I answer every serious message.

Get in touch
Email
hello@farajai.com
GitHub
@FlamuxDev
Based in
Amman, Jordan (GMT+3)