Muhammad Ahsan · Riyadh, Saudi Arabia

AIEngineer

open to Forward Deployed Engineer roles in Saudi Arabia + remote

My approach: start with the people who own a workflow, then design, ship and harden the multi-agent system that runs it — LangGraph, CrewAI and MCP, with a human approval step where it matters.

See how I deploy

View work

Autonomy isn't prompted. It's engineered — part by part.

Scroll to orchestrate

How I deploy

From a real workflow to a guarded agent in production.

The pattern behind the agent workflows I built at Wanhar Impex and the WhatsApp assistant in Cliniwo.

Discover
Map the real workflow, the people in it and the data it touches.
Orchestrate
A LangGraph or CrewAI orchestrator splits the job across specialist agents.
Ground
RAG over the documents the workflow already uses, and MCP tools wired to the systems it runs on.
Guard
Scope limits, guardrails and refusal tests before anything ships.
Approve
A person signs off anything binding before it is committed.
Deploy & observe
Production with logs, evals and cost tracking; lessons flow back to the orchestrator.

Agent Command Center

The orchestration pattern I run in production. This isn't a mockup — it's the real architecture pattern behind my production multi-agent systems.

Orchestration graph

hover · tap · focus a node

RUN 042orchestrator@ahsan · livestreaming

SYSsystem orchestrator online · 6 workers synced · guardrails=strict

HANDOFForchestrator new goal received → match vendor for shipment #A-4471

THINKplanner decomposing goal into 4 steps: retrieve · draft · review · execute

HANDOFFplanner → researcher: gather qualified vendors + freight rates

TOOLresearcher vector_search(query='HS-code 7208 suppliers, KSA lane') → 12 hits

RESULTresearcher ranked 3 vendors by rate, lead-time, reliability score

Framework arsenal

The orchestration stack I build with — each grounded in real projects and certifications.

16 frameworks

LangGraph

ExpertOrchestration

Built browser-integrated workflow sidekicks and the autonomous agent loops that automate client comms and vendor matching at Wanhar Impex.

CrewAI

ExpertOrchestration

Engineered autonomous agent teams that draft contracts and coordinate multi-step business workflows.

AutoGen

AdvancedOrchestration

Designed multi-agent conversation networks and swarms for collaborative task decomposition.

Claude Agent Teams

AdvancedOrchestration

Deployed multi-agent networks and orchestrators with Claude Code subagents, driving automated feature development wired into Jira and GitHub.

OpenAI Agents SDK

AdvancedOrchestration

Orchestrated complex multi-agent systems and hand-off logic across specialized worker agents.

Bedrock AgentCore

AdvancedOrchestration

Built production multi-agent loops on AWS with Lambda, SQS, and Aurora Serverless for scale.

Model Context Protocol

AdvancedProtocol

Integrated multi-server MCP environments so agents reach tools, data, and each other through a shared protocol.

LangChain

AdvancedLLM Engineering

Composed RAG chains and production interfaces over open-source vector datastores.

LangFuse

ProficientObservability

Instrumented LLM observability, cost tracking, and guardrails across production agent runs.

Hugging Face

AdvancedLLM Engineering

Fine-tuned and served open-source models; ran QLoRA on top of the HF ecosystem.

Claude Code

ExpertTooling

Runs high-velocity delivery with Claude Code, Cursor, and MCP — context managed via AGENTS.md and CLAUDE.md.

Token & Context Optimization

AdvancedOptimization

Cut LLM cost and latency with prompt/context compression, caching, and token budgeting — measured and tuned through LangFuse.

LLMOps

AdvancedObservability

Run production LLM operations end to end — observability, cost tracking, guardrails, and evals — alongside classic MLOps.

Agent Deployment

AdvancedDeployment

Ship agents to production on AWS with Bedrock AgentCore, Lambda, SQS, and Aurora — automated via Terraform and GitHub Actions.

GSD

AdvancedMethodology

My agentic workflow methodology for shipping features fast with coordinated coding-agent teams.

Gastown

AdvancedOrchestration

Used Gastown to build and wire multi-agent workflows.

Autonomous Vendor-Matching Pipeline

Wanhar Impex · Director & AI Engineering Lead

real workflow · condensed

01
Intake
Orchestrator
Shipment request enters the graph as typed state.
02
Plan
Planner
Goal decomposed into retrieve → draft → review → execute.
03
Retrieve
Researcher
RAG over the vendor + freight vector store.
04
Draft
Coder
Bilingual RFQ and contract terms generated from context.
05
Audit
Reviewer
Security & privacy guardrails redact and approve.
06
Execute
Executor
Approved action dispatched through MCP tool servers.
07
Trace
Observer
Tokens, cost, and latency logged via LangFuse.

A real production workflow — condensed. Automates vendor sourcing, outreach, and contract drafting with a security gate on every hand-off.

Selected Work

All projects ↗

01 · Agentic AI

SaaS · WhatsApp-native · Arabic / English

Cliniwo — WhatsApp AI Clinic Platform

A multi-tenant SaaS where a WhatsApp assistant answers patients in Arabic and English and books into the clinic's real calendar. In pilot at one clinic in Riyadh.

Next.js 16TypeScriptSupabase (Postgres + RLS)WhatsApp Cloud API

02 · Agentic AI

Voice / Chat · Privacy-First

Local "Jarvis" AI Assistant

A personal voice and chat assistant in Python, designed for local LLMs; the current code calls hosted models through OpenRouter.

PythonLlamaGemmaOpenRouter

03 · Agentic AI

Multi-Agent · Emergent Behavior

Hybrid Intelligence Simulation Game

A civilization sim where NPCs reason with local LLMs and remember via SQLite.

PygameLocal LLMsSQLiteMulti-Agent

04 · Agentic AI

Automation · B2B Lead Generation

Google Maps Lead Scraper

A high-performance scraper automating B2B lead generation and business auditing.

PlaywrightPythonWeb ScrapingLead Generation

05 · AI / ML

Classification & Retention

Customer Churn Prediction

An ML pipeline predicting subscriber churn to power retention campaigns.

Random ForestXGBoostSMOTEScikit-learn

06 · RF Security

RF Security Research · Controlled Test

Car Keyless Entry Replay (HackRF)

A replay study of a vehicle's keyless-entry signal using a HackRF One.

HackRF OneURHDragonOSRKE Protocol

Capabilities

Agentic AI & Orchestration

Multi-agent networks, swarms, and autonomous loops in production.

LangGraphCrewAIAutoGenOpenAI Agents SDKBedrock AgentCoreClaude Agent TeamsMCP

LLM Engineering

From inference to training — RAG, fine-tuning, and vector stores.

RAG PipelinesQLoRA Fine-TuningVector DatastoresLangChainHugging FaceGradioPrompt Compression

MLOps & Cloud

Serverless AI on AWS, Azure, and GCP — automated with IaC.

AWS Lambda / SQSAurora ServerlessTerraformGitHub ActionsLangFuseCI/CD

Machine Learning & Data

Classical ML, deep learning, and statistical modeling.

PythonScikit-learnTensorFlow / KerasXGBoostPandasNLP / NLTKComputer VisionTime Series

Dev Automation

High-velocity delivery with agentic coding tools.

Claude CodeCursorCopilotCodexAGENTS.md / CLAUDE.mdJiraGitHub

Certifications earned

Including Google Advanced Data Analytics, Python 3 Specialization, EHE, and Agentic AI tracks.

Cybersecurity certificates

EC-Council Ethical Hacking Essentials (EHE), Infosec Applied ChatGPT for Cybersecurity, and TryHackMe Pre Security and Introduction to Cyber Security.

Engineering case studies

Across agentic AI, machine learning, live web and SaaS products, games, RF security, and IoT systems.

5+ yrs

Shipping production systems

Leading AI, automation, and software delivery at Wanhar Impex, 2020–2026.

Services

What I can build for you.

Production-grade agentic engineering: secure multi-agent systems, RAG over your private data, and custom LLM integrations — grounded, observable, and deployed serverless on AWS, Azure, or GCP.

All services ↗

Agentic AI Consulting

Strategy and architecture for autonomous, multi-agent systems that deliver commercial outcomes.

↗

AI Automation

Autonomous workflows that replace manual, repetitive operations end to end.

↗

Custom Agent & Chatbot Development

Bespoke agents, copilots, and chatbots — grounded, secure, and observable.

↗

AI Cost & Token Optimization

Cut your LLM bill without cutting quality — context engineering, caching, and the right model for each job.

↗

Credentials · 40 earned

40 documented certifications across agentic AI, data science, security, and engineering — the foundation under the work.

Ed Donner · Proficient AI Engineer · Aug 2026
Udemy · Jul 2025
Udemy · Apr 2026
Udemy · Jun 2026
Udemy · Jun 2026
Udemy · Apr 2026
Mentoga · Jun 2025

Supporting Skills

The foundations under the AI — security, data, and hardware.

Cybersecurity

Ethical hacking, forensic analysis, and secure-by-design pipelines.

Ethical Hacking (EHE)TryHackMeWeb App TestingDigital ForensicsSecurity Guardrails

Data Analytics

Google-certified advanced analytics and statistical modeling.

Regression AnalysisStatisticsTableauEDAData Pipelines

IoT & Wireless SDR

RF communication, SDR, and physical sensor integration.

HackRF / SDRRTL-SDRArduinoRaspberry PiESP8266Embedded C++Flutter

Open to Work

Available for AI Engineer, Generative AI Engineer, Machine Learning Engineer roles — on-site in Saudi Arabia or remote worldwide.

Start a conversation

Ask my AI twin

● Actively open to work

AIEngineer

From a real workflow to a guarded agent in production.

Discover

Orchestrate

Ground

Guard

Approve

Deploy & observe