Senior Full-Stack AI Engineer / Systems Architect

I build production AI systems end‑to‑end — from infrastructure to model internals to shipped product.

Agentic AI, RAG, and computer vision — plus the backend, infrastructure, and MLOps to ship them. I own the whole arc, solo or leading a small team. Based in Morocco, working worldwide.

Get in touch See the work

~/ the stack · projection

L4Products at scaleNext.js/React · FastAPI · Docker/Kubernetes — platforms serving 5,000+ users

L3AI systemsagentic RAG (LangGraph/LangChain) · Neo4j knowledge graphs · edge AI (NVIDIA Jetson) · MLOps

L2Models & pipelinesPyTorch · TensorFlow · Hugging Face · YOLO/Detectron · GPT/BERT/Llama fine-tuning

L1ML internalsperceptron · MLP · backprop · Transformer — by hand

L0Machine & systemsC · C++ · assembly · shells · HTTP servers from scratch

20×

User throughput scaled

~90%

Grounded RAG accuracy

5,000+

Users served on a platform I built

~30%

Lower AI / inference spend

Worked with

UM6PSURETREEPROD'AIRDECENTUREIRESEN42 Network / 1337OCPEDFLeytonCognitixVanguardGEPUM6PSURETREEPROD'AIRDECENTUREIRESEN42 Network / 1337OCPEDFLeytonCognitixVanguardGEP

What I do

Five ways I turn AI into something you can ship.

Most people wire AI APIs together. I architect the whole system — and I've built the internals by hand, so I can reason about what's happening underneath.

AI Agents & RAG Systems

Autonomous agents and retrieval grounded in your data — knowledge graphs, vector search, document processing, and reporting that reflects your business.

Applied ML & Computer Vision

From model selection to deployment, including real-time vision on edge hardware (NVIDIA Jetson / YOLO), OCR pipelines, and predictive models.

AI Product Engineering

Turning prototypes into scalable products — FastAPI/Python backends, React/Next.js frontends, PostgreSQL/Redis, containerized and deployed.

MLOps & Cloud Infrastructure

Reproducible training and deployment pipelines, CI/CD, Docker/Kubernetes, and cost optimization for LLM and inference workloads.

Fractional AI Lead / Advisory

Architecture reviews, technical standards, roadmap, and mentoring — the value of a senior architect without a full-time hire.

→

Not sure which you need?

Tell me the problem and I'll tell you the shortest path to shipped.

Selected work

Systems I architected, built, and shipped.

Each one: the real problem, the architecture decision that mattered, and what it produced.

AI Engineer (Consultant) · Fintech · Remote, Dubai · 2025

Agentic RAG grounded in a knowledge graph

Agentic AI · RAG

The problem

A fintech team needed bespoke GenAI reliable enough for internal teams and external clients — but LLMs alone hallucinate and don't know company-specific facts.

Architecture

Built agentic workflows with LangGraph + LangChain on a FastAPI backend for real-time insight gathering and decision support.

Grounded agents in a Neo4j knowledge graph + vector databases so answers traced back to source.

Integrated outputs into existing tooling via HubSpot.

Outcomes

~90% answer accuracy

~60% less manual reporting time

~30% lower LLM spend

5 use cases into production

LangGraphLangChainFastAPINeo4jVector DBHubSpot

ProvesProduction agentic AI that's grounded, integrated, and cost-aware — not a demo.

Senior R&D Engineer / Technical Lead · UM6P (SCI) · 2025–present

Re-architecting an LLM platform to scale 20×

LLM Platform · Lead

The problem

Deliberatorium used LLMs to structure large-scale deliberation but existed as research code. I led its rebuild into a production system (Aideator).

Architecture

Re-architected the full stack: React/TypeScript · FastAPI · PostgreSQL · Redis.

Containerized with Docker Compose, deployed on DigitalOcean with real-time incident detection.

Owned the legacy migration, preserving historical data while re-implementing research concepts as production LLM features.

Set the team's engineering standards (deploys, code review, CI).

Outcomes

20× capacity (~50 → 1,000+ concurrent users)

5 engineers led

zero historical data lost in migration

ReactTypeScriptFastAPIPostgreSQLRedisDocker

ProvesTechnical leadership and architecture under real constraints.

AI Engineer · PROD'AIR · Marrakech · 2024–2025

Real-time computer vision on the edge

Computer Vision · Edge AI

The problem

A smart-city digital twin needed real-time object detection. Cloud inference was too slow, too costly, and not accurate enough.

Architecture

Built real-time YOLO inference on NVIDIA Jetson to detect and classify urban objects.

Moved inference from cloud to the edge — one decision, three gains.

Built an energy pipeline across 50+ IoT sensors to surface leaks and anomalies.

Delivered Llama-2 RAG with OCR for employee self-service.

Outcomes

~60% better detection accuracy

~30% lower cloud costs

~50% faster deployment

50+ IoT sensors

YOLOJetsonEdge AIIoTLlama-2OCR

ProvesEnd-to-end ownership of a hard, physical-world AI system.

Experience — from founder to technical lead.

Oct 2025 – Present

Senior R&D Engineer / Technical Lead

UM6P (SCI), Rabat

Leading a 5-person team rebuilding an LLM deliberation platform into production (Aideator); owning architecture, the legacy migration, and engineering standards.

Jun 2025 – Oct 2025

AI Engineer Consultant

SURETREE, Remote (Dubai)

Core contributor on a fintech AI/Data team shipping bespoke GenAI/RAG — agentic systems, a Neo4j knowledge graph, HubSpot integration, ~90% answer accuracy.

Nov 2024 – Apr 2025

AI Engineer

PROD'AIR, Marrakech

Architect and lead developer of a real-time CV system on NVIDIA Jetson for a smart-city digital twin, plus a 50+ sensor IoT energy pipeline and Llama-2 RAG.

May 2023 – Jun 2024

R&D Engineer

DECENTURE, Remote (London)

Built 5+ production NLP models (GPT/BERT) at ~95% validation accuracy and reproducible MLOps, plus full-stack features across a web/mobile product.

Jun 2021 – May 2023

Co-Founder & Technical Lead

PHOENIX SERVICES, Benguerir

Co-founded a campus-digitalization startup, led it from POC to ~5,000 users, and ran the venture — fundraising, partnerships, hiring, product.

2021 – 2022

Full-Stack Developer

Upwork & IRESEN

AI / ML

LLMs & fine-tuningLangGraphLangChainRAGNLP & TransformersYOLODetectronMediaPipeOCR

ML Engineering & MLOps

PyTorchTensorFlowHugging Facescikit-learnOpenAI APIAnthropic APIEdge AI (Jetson)IoT pipelines

Backend & Full-Stack

FastAPIDjangoFlaskNode.jsReactNext.jsStreamlitREST APIs

Data & Databases

PostgreSQLMySQLRedisNeo4jVector DBsETLPandas / NumPy

Cloud & DevOps

DockerKubernetesCI/CDAWSDigitalOceanLinuxGit

Languages

PythonTypeScriptJavaScriptCC++RustSQLBash

Credentials

Education & recognition

exp → 4+ yrs shipping AI & full-stack

langs → English · French · Arabic

edu → RNCP7 (Master's-equiv.)

also → ex-founder · club president

I'm Elmahdi Elaazmi, a full-stack AI engineer and systems architect based in Morocco, working with clients worldwide.

I take AI from prototype to production and own the whole arc — architecture, models, backend, infrastructure, and delivery. Over the past 4+ years I've shipped agentic RAG for fintech, real-time computer vision on edge hardware, production NLP for ed-tech, and an LLM platform that scaled 20× — sometimes solo, sometimes leading a small team.

What sets me apart is depth. Through a rigorous engineering track, I built neural networks, a Transformer, HTTP servers, and container tooling from scratch. So I don't just wire AI APIs together — I understand what's happening underneath.

I also think like a founder, because I've been one: I co-founded a campus-digitalization startup that reached ~5,000 users. That ownership shows up in every engagement — I care about the outcome, not just the ticket.

References

What people I've worked with say.

“He didn't just implement what we asked for — he pushed back on the architecture until it could actually handle our scale. That's rare.”

— Placeholder, Engineering Lead, Fintech client

“We handed him an ambiguous, half-defined problem and got back a production system. He owns outcomes, not tickets.”

— Placeholder, Technical Director, Research institution

Client references available on request.

Let's build

Have an AI idea that needs to become real?

Whether it's a prototype that needs to scale, an agent that needs grounding, or a system that needs an architect — tell me the problem and I'll tell you the shortest path to shipped.

Email me →Get in touch on WhatsApp Hire me on Upwork

elaazmielmahdi@gmail.com LinkedIn ↗GitHub ↗+212 675 874 185