Hi, I'm Burak.
I lead the open source team at Türkiye's Presidency of Migration Management. With my team I run its Linux infrastructure, build open source applications and fine-tune the Bilge8 language models.
My work Open source contributions Contact
Scroll down
I shared my first code in 2006.
I put three small scripts on ASP İndir (opens in a new tab), a Turkish code-sharing site. One of them tried to block flood attacks on websites. People commented that DDoS isn't that easy to solve. They were right.
In 2009, at 19 and still a student at Adnan Menderes University, I built e-webOS, a virtual computer that opened in a web browser. I started from the open source eyeOS, reworked its skeleton and wired OpenOffice into it. Word, Excel and PowerPoint files opened with nothing installed, even from a phone. I ran it on my own servers in a data centre in İzmir. It even made the news. (opens in a new tab)
Years later I did the same thing again with Bulut: took an open source project and grew it to fit what we needed.
My first job was building university networks.
If you've ever connected to eduroam at a Turkish university, there's a chance I set it up. We also built a network for ISAF at Mazar-i-Sharif Airport in Afghanistan.
Then I took over the Linux servers at Hacettepe University, and I haven't left Linux since.
Then came a fingerprint system for Bhutan.
At Papilon I looked after the servers behind biometric systems. First we installed a ballistic identification system for the criminal police laboratories in Sudan. The year after, the Royal Bhutan Police project came along: a palm and fingerprint identification system for the whole country.
I designed and deployed it. In November 2016 they gave me a plaque of appreciation.
Then came the hospitals.
At KEYDATA I worked on the infrastructure of the Kayseri, Konya and Manisa city hospitals. There were more than 600 servers, and since a hospital never closes, the systems couldn't go down either.
We built everything on open source; there was no Windows. The desktops ran Linux too: KeyOS, developed by my dear friend Mehmet Yılmaz. The servers were mine, the desktops were Mehmet's.
Building open source services at the Presidency.
I've been at the Presidency of Migration Management since 2021. For the first three years I was on the staff of a UNHCR project there, and became team lead in June 2023. Since September 2024 I've continued as team lead through Türksat. What we started under the name Dijital'e GÖÇ has grown into nine applications. They all run on the Presidency's own servers, and the data never leaves. GöçBil is a team project; I develop the others and adapt them to the Presidency's needs.
I set up the infrastructure underneath as well: more than 260 virtual machines on Proxmox, Kubernetes clusters, and databases that stay up when a server goes down.
The people who keep all of this running every day are my team. None of it would work without them.
- Bulut (opens in a new tab) File sharing and shared document editing, with 18+ modules
- Kırlangıç (opens in a new tab) Messaging, voice and video calls
- Konferans (opens in a new tab) Video conferencing and remote training
- Anket (opens in a new tab) Building surveys and analysing the results
- Analitik (opens in a new tab) Website visitor statistics
- GöçBil (opens in a new tab) A team project; I build the services that run its face recognition and anti-spoofing models, and support the development team
They are all at acikkaynak.goc.gov.tr (opens in a new tab).
I try to give back what I take.
When I find a bug in the open source software we use, I fix it and send the fix upstream. As of November 2026, 58 of my pull requests (opens in a new tab) have been merged across 20 open source repositories.
- Nextcloud Talk (opens in a new tab) In clustered setups people were dropped from calls. I traced the cause and wrote the test; the fix was merged in November 2026
- Talk live captions (opens in a new tab) Video decoded for nothing, a language setting leaking between rooms, captions that never started; all five fixes were merged
- vLLM (opens in a new tab) A tuned config for the NVIDIA A100; kernel time fell by 2.5% to 19.4%
- GPUStack (opens in a new tab) Requests weren't spread evenly across model replicas; now they take turns evenly
- Dify (opens in a new tab) The translation check missed broken placeholders; it now catches them in every language
They range from security fixes and crash fixes to interface translations. The rest is on GitHub (opens in a new tab).
From hardware to the language model: the whole stack.
I build infrastructure from the bottom up: hardware at the base, a language model at the top. These are things I have set up and run over the years. Tap a layer to see what's in it.
-
Hardware Physical servers, GPU servers, Switches and routers 6
Physical servers · GPU servers · Storage systems · Switches and routers · Firewalls · Wireless access points
-
Operating systems Debian, Ubuntu, Oracle Linux 8
Debian · Ubuntu · Pardus · Red Hat · Oracle Linux · Rocky Linux · Alma Linux · Windows Server
-
Storage Ceph, GlusterFS, OpenZFS 4
Ceph · GlusterFS · OpenZFS · TrueNAS
-
Virtualisation and containers Proxmox VE, Kubernetes, Docker 9
Proxmox VE · Proxmox Backup Server · KVM · VMware vSphere · OpenStack · Docker · Kubernetes · KubeSphere · Rancher
-
Network and services HAProxy, Nginx, FreeIPA 8
Nginx · Apache · HAProxy · Keepalived · BIND · KEA DHCP · FreeIPA · Active Directory
-
Databases PostgreSQL, Patroni, MariaDB 10
PostgreSQL · Patroni · MariaDB · Galera Cluster · MongoDB · Redis · etcd · pgvector · Qdrant · Milvus
-
Automation and management Ansible, AWX, Oracle Linux Manager 5
Ansible · AWX · Oracle Linux Manager · FOG Project · GLPI
-
Monitoring and security Zabbix, Grafana, Wazuh 6
Zabbix · Prometheus · Grafana · Elasticsearch · OpenSearch · Wazuh
-
Email and collaboration Nextcloud, BigBlueButton, Zimbra 5
Postfix · Zimbra · Nextcloud · BigBlueButton · Jitsi
-
AI infrastructure GPUStack, vLLM, MS-SWIFT 12
MS-SWIFT · LLaMA-Factory · vLLM · SGLang · TensorRT-LLM · Triton Inference Server · GPUStack · LiteLLM · LangChain · LangGraph · LlamaIndex · Langfuse
These days I fine-tune language models.
You ask Göç Uzmanı a question about migration law, and it shows you the legal text behind its answer. Behind it is Bilge8 (opens in a new tab), our family of language models, running on the Presidency's own servers.
I fine-tune the Bilge8 models and am the principal investigator of our two projects, on TRUBA and on MareNostrum 5 in Barcelona. We don't train the models from scratch; we take open models and keep training them on data from our own field. That is called fine-tuning, and the result is a domain-specific model.
Turn the 8 in its name on its side and you get the infinity sign. We put it there so the learning never stops.
Outside work I'm also building models of my own: Cezeri and Harezmi.
Say hello.
If you have a question, or just want to say hi, write to me. The language school I attended in Canada sent me home with its "Friendliest Student" award, so don't be shy.
Write to me on LinkedIn (opens in a new tab)
LinkedIn (opens in a new tab) GitHub (opens in a new tab) Hugging Face (opens in a new tab)
Soon a language model I'm working on will answer your questions here too.