Skip to content
Mahdi Talal
Hire me
Menu

(01) AI Engineer · Empire Media

Beirut, Lebanon

Mahdi Talal – Freelance AI Engineer

Portrait of Mahdi Talal

AI EngineerAutonomous agentic coding pipelines & production LLM systems

I'm Mahdi Talal, an AI Engineer at Empire Media in Beirut, Lebanon. I build autonomous AI coding pipelines (one has opened 200+ pull requests) and production MCP servers, including a multi-tenant Google Workspace MCP server used by 15 people.

At a glance

Role
AI Engineer
Employer
Empire Media
Location
Beirut, LebanonEET/EEST (UTC+2/+3)
Current workplace
Remote
Building since
2023
Languages
Arabic (native), English (professional)

Available for freelance AI projects, and open to the right full-time role · Beirut, Lebanon

(02) Proof

By the numbers

Each figure shows its source and the month it applies to.

pull requests openedAutonomous AI Coding Pipeline
200+
Source: My count from the pipeline's GitHub history · Sep 2026
tools across 12 Google servicesGoogle Workspace MCP Server
186
Source: CAPABILITIES.md in the server repository · Sep 2026
people using itGoogle Workspace MCP Server
15
Source: My own count · Sep 2026
daily.dev MCP toolsRemote MCP Servers: daily.dev & fal.ai
42
Source: lib/tools.ts and the live /health endpoint · Sep 2026

(03) Flagship work

Selected work

  1. Illustration: an isometric 3D software factory where cobalt code blocks ride a conveyor through three scanner gates, past a small robotic arm, to a final station topped by a glowing marigold approval light
    Employment project

    Empire Media · Architect and sole builder · May 2026 – Present

    Autonomous AI Coding Pipeline

    Label a GitHub issue; an agent plans, writes tests, codes and opens a pull request that gated AI reviewers check.

    200+Pull requests opened

    Highlights

    • Issue-to-PR authoring with a spec-refinement step and up to two clarification rounds instead of guessing.
    • Guardrail hooks the agent cannot edit: a TDD guard, a production-push block, and sensitive-path and credential-egress guards.
    • Bounded loops: capped iterations, stuck detection and a retry ladder before a needs-human label.
    • A juror subagent filters review findings by confidence; security findings are never demoted.
    • A weekly retro proposes memory updates as human-reviewed pull requests.
    • Claude Code
    • Claude API
    • GitHub Actions
    • Python
    • TypeScript
    • Model Context Protocol
    • Playwright
    • +3 more

    Read the Autonomous AI Coding Pipeline case study

  2. Illustration: an isometric 3D navy server tower cabled to four separate glass pods, each with its own marigold padlock and holding a mail, calendar or document object
    Employment project

    Empire Media · Sole developer · Jun 2026 – Present

    Google Workspace MCP Server

    One Claude connector that gives each employee isolated access to their own Google accounts.

    186Tools across 12 Google services

    Highlights

    • 186 tools across Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Chat, Contacts, Tasks, GA4 and Search Console.
    • An OAuth 2.1 authorization server with PKCE and dynamic client registration, built into the MCP server.
    • Per-user workspaces with encrypted refresh tokens; every tool call is scoped to one linked Google account.
    • TypeScript
    • Node.js
    • Fastify
    • Model Context Protocol
    • PostgreSQL
    • Google Workspace APIs
    • Docker
    • Railway
    • +4 more

    Read the Google Workspace MCP Server case study

  3. Illustration: an isometric 3D workshop where blank paper documents travel along a conveyor through a scanner, a press and a checking arch, and come out stamped with marigold seals
    Personal project

    Founder and sole developer · Feb 2026 – May 2026

    CVForge AI

    AI career platform: RAG-tailored CVs and cover letters, ATS scoring, application tracking and mock interviews.

    9BullMQ worker types

    Highlights

    • Retrieval-augmented CV and cover-letter generation with ATS scoring, version diffs and PDF export.
    • A Kanban application tracker with URL scraping, CSV bulk import and per-application AI chat.
    • Mock interviews across 7 interview types and 4 difficulty levels.
    • Next.js
    • TypeScript
    • PostgreSQL
    • pgvector
    • Prisma
    • Redis
    • BullMQ
    • Stripe
    • OpenAI API
    • Docker
    • GitHub Actions
    • +7 more
  4. Illustration: an isometric 3D city block with a dispatch hub, a neat queue of delivery scooters, glowing cobalt route lines along the streets and a marigold chat bubble above the lead scooter
    Employment project

    Loadless.ai · for Quicky · Lead developer · Jan 2026 – Apr 2026

    Quicky Delivery Dispatch ERP

    Dispatch ERP for a delivery company: fair driver queue, automated WhatsApp dispatch, commissions and payments.

    5Driver-selection levels

    Highlights

    • A fair FIFO driver queue with a five-level selection fallback, from manual assignment to stacking on the longest-busy driver.
    • Automated Arabic WhatsApp messages to drivers and vendors through n8n and Respond.io webhooks, with a one-click retry when a send fails.
    • Driver shifts, commission snapshots, a payments ledger and running balances for finance roles.
    • Five roles with role-scoped data windows, so each role sees only the orders it needs.
    • Rebuilt from an n8n and Google Sheets prototype into a containerised web app deployed with GitHub Actions.
    • Next.js
    • JavaScript
    • TypeScript
    • PostgreSQL
    • Prisma
    • n8n
    • WhatsApp
    • Docker
    • Caddy
    • GitHub Actions
    • +6 more

See every project and case study

(04) About

Experience

  1. NowApr 2026 – PresentFull-time

    AI Engineer

    Empire Media logoEmpire Media (Viberant Group of Companies) · Remote

    Builds the company's AI platform: an autonomous AI coding pipeline, production MCP servers and agent operations for marketing, accounting and HR teams.

    • Designed and built an autonomous AI coding pipeline that turns labelled GitHub issues into tested pull requests behind gated AI review; it has opened 200+ pull requests.
    • Built a multi-tenant Google Workspace MCP server with its own OAuth 2.1 authorization server: 186 tools across 12 services, used by 15 people.
    • Built the agent-operations layer of the ApplyBoost marketing dashboard, where scheduled Claude agents report runs and raise decisions for people to approve.
    • Packaged Claude Code skills, plugins and hooks into an internal plugin marketplace for non-engineering departments.
    • Built an internal spend dashboard with 17 billing connectors, much of it written through the coding pipeline itself.

  2. Dec 2025 – Apr 2026Full-time

    AI & Software Engineer

    Loadless.ai logoLoadless.ai · Remote

    Built production products for Loadless.ai clients: a delivery dispatch ERP, a multi-tenant booking SaaS and a market-intelligence platform.

    • Built Quicky, a delivery dispatch ERP in production at Quicky: a fair FIFO driver queue, automated WhatsApp dispatch, commissions and payments.
    • Built Chairly, a multi-tenant booking, POS and CRM SaaS for barbershops and clinics with WhatsApp automation and AI hairstyle previews.
    • Built a market-intelligence platform that turns n8n pipelines into branded economic-calendar and news digests for email, WhatsApp, Telegram and Instagram.
    • Shipped containerised deployments with GitHub Actions, Docker and Caddy.

  3. Jul 2025 – Dec 2025Full-time

    AI & Automation Specialist

    Morph Collective logoMorph Collective · Remote

    Built AI products and automations for the new AI division of a Dubai-based digital agency.

    • Led backend development of Pulse, an AEO SaaS that tracks how ChatGPT, Gemini, Claude, Grok and Perplexity mention and cite a brand.
    • Designed Pulse's visibility-scoring model and its Stripe subscription billing with plan limits.
    • Engineered an n8n client sync between Phorest and Respond.io with polling and error handling, plus a salon AI chat concierge.

  4. Jul 2025 – Apr 2026Part-time

    AI & Automation Manager

    Skool logoAmir Discoveries (Skool community) · Remote

    Taught AI tools and automation workflows to a Skool community.

    • Ran structured sessions and live Q&A on AI tools and automation workflows.

  5. Jan 2023 – Jul 2025Freelance

    Full-Stack Web Developer

    Freelancer.com logoSelf-employed (Freelancer.com) · Remote

    Built custom web applications and automations for small businesses.

    • Built full-stack web applications for small businesses, from storefronts to management systems.
    • Built n8n scraping and data-collection workflows.

Education & certificates

(05) Principles

How I build

  1. Agents with hard guardrails

    Deterministic orchestration around LLM workers: bounded loops, hooks the agent can't edit and schema-typed handoffs.

  2. Humans stay in charge

    Agents propose; people merge pull requests, answer clarifications and approve decisions.

  3. Production MCP engineering

    Remote MCP servers with OAuth 2.1, per-tenant isolation and verified side effects.

  4. Spec-first and documented

    PRDs with rejected alternatives, decision records and honest notes on what is verified versus assumed.

Toolbox

AI & agents
Claude API, Claude Code, Model Context Protocol, OpenAI API, LangChain, n8n, pgvector, fal.ai
Languages
TypeScript, JavaScript, Python, PHP, SQL, Dart
Frontend
React, Next.js, Tailwind CSS, Redux, Electron
Backend
Node.js, Express, Fastify, NestJS, Laravel, Socket.IO
Data
PostgreSQL, MySQL, MongoDB, Prisma, Drizzle ORM, Redis, BullMQ, Supabase
DevOps & cloud
Docker, GitHub Actions, Cloudflare, Railway, Vercel, AWS, Caddy
Integrations
Stripe, WhatsApp, Telegram Bot API, Google Workspace APIs
Testing
Vitest, Playwright, Jest

(06) Questions

Common questions

What kind of AI engineering projects do you specialize in?

I focus on production agentic systems: autonomous AI coding pipelines (multi-agent workflows with automated linting, test suites, and pull request generation), production Model Context Protocol (MCP) servers, and enterprise automation connecting LLMs to internal tools and APIs.

How do your autonomous coding pipelines prevent hallucinations and broken code?

Every agent action operates within strict, deterministically enforced guardrails. The pipeline runs automated linters, type checks, unit tests, and security scans in isolated containers before creating a pull request. Humans always review and approve before any code merges to production.

What is the Model Context Protocol (MCP) and how do you use it?

MCP is an open standard that allows LLMs to securely connect to external tools, databases, and enterprise services. I build production-grade, multi-tenant MCP servers—such as a Google Workspace server supporting Gmail, Calendar, and Drive with per-user OAuth 2.1 with PKCE authentication.

Are you available for freelance projects or full-time roles?

Yes. I am currently available for freelance AI engineering projects (autonomous pipelines, MCP server development, and LLM integrations), and I am open to discussing compelling full-time AI engineering opportunities.

(07) Contact

Let's build

Available for freelance AI projects, and open to the right full-time role

Email

hello@mahditalal.com

Elsewhere