Distributed Systems

HPC & GPU

Backend & Platform

Toulouse, France

Anar Mammadli — Portfolio

Hello, I'm

Anar
Mammadli

Solution Architect · Senior Software Engineer
Data Science Researcher

I design and run the systems other software depends on Linux clusters, job schedulers, GPU platforms and the backends that sit on top of them. Seven years running research computing at ASOIU through to 2025, four years and counting on production backends at Markify, and a doctorate on explainability alongside both.

Linux HPC
SLURM
CUDA
Python
Node.js
PHP

Anar Mammadli, arms folded, in a black sweater.

Markify Senior Software Engineer Matrix Academy Deep Learning & Generative AI IRIT Data Science Research

PlatformWork

The infrastructure end. These have no interface to photograph a cluster, a scheduler, a registry pipeline so they are set down rather than shown.

2018 – 2025

HPC Cluster Platform

Azerbaijan State Oil & Industry University

Estate Mixed GPU/CPU Linux cluster on OpenHPC
Scheduling SLURM, PBS Pro, Moab, TORQUE
GPU CUDA, NCCL, MIG partitioning, NVLink
Benchmarks HPL, HPCG, IO500, OSU, IOR, mdtest
Tenure Seven years, engineer through to platform owner

I owned the university's research computing platform end to end for seven years — head, login and compute nodes across a mixed GPU/CPU estate. I designed the scheduling layer that decided who got the machine and when: partitions, preemption, accounting and user quotas across four schedulers. I tuned the MPI and OpenMP build matrices researchers compiled against, ran the benchmark suite that justified hardware spend, and set firmware and driver policy for the NVIDIA platform. The Python and Bash tooling, the job-submission portals and the CI that kept node images reproducible were mine too. The programme wound down in 2025; the scheduling and throughput work is what I carry into platform roles now.

OpenHPC
SLURM / PBS
CUDA / NCCL
MPI
NVLink
Prometheus / Grafana

2022 – Present

Markify

Senior Software Engineer · Stockholm, remote

Domain Global trademark search and watch
Tenure Four years, current role

Four years on the engineering team behind a global trademark service. The work is backend and platform against large, messy, jurisdiction-specific registry data, in a domain where a missed record is a legal problem rather than a rendering bug — so correctness, reconciliation and traceability set the bar rather than throughput alone. Commercial work, so specifics stay inside.

PHP
Backend
Data Pipelines
Remote

SelectedProjects

Products I built and still ship. Three are Azerbaijani-language, where the dataset, the dubbing pipeline or the localisation layer has to be built before anything else can be — and one is an AI pricing tool for SaaS teams.

Anifan

anifan.az
Anime & manga platform

Umanage

umanage.az
Education management SaaS

Yasaq

smthGames · iOS & Android
Azerbaijani word game

Cuprice

cuprice.io
AI pricing for SaaS

What IArchitect

Four things I get called in for, and the evidence for each sits in the section above.

I build systems that can account for their own decisions — the scheduler that can say why it chose that job, the pipeline that can show where a record came from.

Cluster & Scheduler Design

Node topology, partitioning, preemption and accounting policy for shared research estates across SLURM, PBS Pro, Moab and TORQUE.

GPU & Performance

CUDA and NCCL stacks, MIG partitioning, NVLink topology, NUMA and BIOS tuning, and a benchmark practice that survives procurement.

Backend & Product Engineering

Multi-tenant SaaS and content platforms in PHP and Node.js — Markify, Umanage and Anifan — with the reconciliation and reporting layers that make their data trustworthy.

Applied ML, NLP & Generative AI

Explainable classification and knowledge-graph generation, serious games and AR/VR interfaces from the doctoral work, and the LLM, RAG and agent material I teach at Matrix Academy.

What ITeach

Architecture is mostly explaining a system to people who don't already understand it. I do that on a schedule.

Deep Learning & Generative AI

Matrix Academy, 2025–present. Prompt engineering, LLM applications, RAG systems, agents and workflow automation.

Data Analytics

Oracle SQL, Power BI and Python, taught to cohorts through to examination on real interview-style problems.

Doctoral Research

ASOIU, with a research affiliation at IRIT Toulouse. Supervisor: Elviz Ismayilov.

Conference Presenting

Presented in person at KES 2024 in Seville and at MaCoSEP 2025.

PublishedWork

Doctoral work on making classifiers explain themselves — the same instinct as building systems that can account for their own decisions.

NLP
Deep Learning
Serious Games
AR / VR
Generative AI

2025

Enhancing User Interaction through Augmented Reality Interfaces

MaCoSEP 2025
with Elviz Ismayilov

2024

Explainability of text classification through ontology-driven analysis in Serious Games

KES 2024 · Seville
with Elviz Ismayilov, Cecilia Zanni-Merk

In progress

Morphology-aware taboo word generation for Azerbaijani and Uzbek

Working paper · not yet submitted
using the Yasaq dataset

Let's TalkSystems

Open to architecture and senior platform roles, and to freelance alongside them. Remote across Europe, based in Toulouse.

The interesting problems are the ones where the hard part is the system, not the screen.

Scheduling, throughput, correctness at scale, and the tooling that keeps a platform legible to the people who inherit it. If that's the shape of your problem, I'd like to hear about it.

Start a conversation

          LinkedIn
          linkedin.com/in/anar-mammadli
        
          Medium
          medium.com/@anarmammadli
        
          Instagram
          instagram.com/anarmammadli
        
          Email
          me@anarmammadli.com
        
          Working
          Remote across Europe · open to freelance
        
          Services
          Solution architecture · platform & IT consulting

Anar Mammadli — Portfolio PORTFOLIO