Onuralp SEZER

@onuralpszr ·Samsun,Turkey ·gravatar.com/onuralpsezer

Senior ML&CV Engineer - Fedora KDE-SIG

311 repos
748 followers
720 following

Kotlin 23%
Python 23%
Rust 14%
C++ 14%
Jupyter Notebook 14%

👋 Greetings Everyone! I'm Onuralp, a seasoned Senior Machine Learning Engineer @Ultralytics with expertise in Python 🐍, Kotlin 📱, C++ ⚙️, and Rust 🦀. My passion lies in the dynamic…

👋 Greetings Everyone!

I'm Onuralp, a seasoned Senior Machine Learning Engineer @Ultralytics with expertise in Python 🐍, Kotlin 📱, C++ ⚙️, and Rust 🦀. My passion lies in the dynamic fields of Computer Vision 👁️, Machine Learning 🤖, and Deep Learning 🧠. Beyond development, I have a strong foundation in DevOps and MLOps, ensuring seamless deployment, automation, and scalability of cutting-edge solutions 🚀.

🧠 `model.summary()`

python
import torch.nn as nn

class Onuralp(nn.Module):
    def __init__(self):
        super().__init__()
        self.role = &quot;Senior Machine Learning Engineer @Ultralytics&quot;
        self.focus = [&quot;Computer Vision 👁️&quot;, &quot;Deep Learning 🧠&quot;, &quot;MLOps ⚙️&quot;, &quot;GPU Kernels ⚡&quot;]
        self.languages = [&quot;Python 🐍&quot;, &quot;Kotlin 📱&quot;, &quot;C++ ⚙️&quot;, &quot;Rust 🦀&quot;, &quot;CUDA PTX ⚡&quot;]
        self.community = [&quot;Fedora Project Contributor 🐧&quot;, &quot;GDG Samsun Organizer 🎤&quot;]

    def forward(self, coffee):
        code = self.write_code(coffee)
        models = self.train_yolo(code)
        return self.deploy_to_edge(models)

🚀 `inference_pipeline.run()`

I specialize in building, optimizing, and deploying state-of-the-art Computer Vision models. From training YOLO architectures at Ultralytics to optimizing inference with ONNX, TensorRT, and ExecuTorch, I bridge the gap between research and production.

🔭 Currently working on: Pushing the boundaries of real-time object detection and tracking at Ultralytics.
🌱 Currently exploring: Vision-Language Models (VLMs), Edge AI, Rust-based ML inference, and GPU kernel programming with cuda-oxide.
💬 Ask me about: YOLO, PyTorch, Computer Vision, Open Source, and Fedora Linux.

🧬 `model.fit(epochs=1000)`

| Epoch 1 | Epoch 50 | Epoch 100 |
|:---:|:---:|:---:|
| 🔲🔳🔲🔳🔳🔲🔳🔲🔳🔲🔳🔲🔲🔳🔲🔳🔲🔳🔲🔳🔳🔲🔳🔲🔳🔲🔳🔲🔲🔳🔲🔳🔲🔳🔲🔳🔳🔲🔳🔲🔳🔲🔳🔲🔲🔳🔲🔳🔲🔳🔲🔳🔳🔲🔳🔲🔳🔲🔳🔲🔲🔳🔲🔳 | ⬛⬛⬛🟪🟪⬛⬛⬛⬛⬛🟪🟪🟪🟪⬛⬛⬛🟪🟪🟪🟪🟪🟪⬛⬛🟪🟪🟪🟪🟪🟪⬛⬛⬛🟪🟪🟪🟪⬛⬛⬛⬛⬛🟪🟪⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ | 🟥🟥🟥🟥🟥🟥🟥🟥🟥⬛⬛⬛⬛⬛⬛🟥🟥⬛🟦⬛⬛🟦⬛🟥🟥⬛⬛⬛⬛⬛⬛🟥🟥⬛⬛🟦🟦⬛⬛🟥🟥⬛⬛⬛⬛⬛⬛🟥🟥🟥🟥🟥🟥🟥🟥🟥⬛⬛⬛⬛⬛⬛⬛⬛ |
| *Raw Data (Noise)* | *Feature Extraction* | *YOLO Object Detected!* |

console
[Epoch 001] 🟢⚪⚪⚪⚪⚪⚪⚪⚪⚪ (Learning Python &amp; C++)
[Epoch 050] 🟢🟢🟢⚪⚪⚪⚪⚪⚪⚪ (Mastering Computer Vision &amp; OpenCV)
[Epoch 100] 🟢🟢🟢🟢🟢⚪⚪⚪⚪⚪ (Training YOLO &amp; Deep Learning Models)
[Epoch 500] 🟢🟢🟢🟢🟢🟢🟢⚪⚪⚪ (Optimizing with ONNX, TensorRT &amp; Rust)
[Epoch 999] 🟢🟢🟢🟢🟢🟢🟢🟢🟢🟢 (Deploying AI @ Ultralytics 🚀)

&gt; Model &#39;Onuralp&#39; successfully converged.
&gt; Current State: Senior ML Engineer Lv2
&gt; Next Objective: AGI &amp; Beyond...

🐧 Open Source & Community

I am a long-term Contributor and Mindshare Member, Fedora KDE SIG Member, Fedora Website and Apps Team Member, and RPM Packager at the Fedora Project. As an RPM packager, I actively maintain and package Python, KDE, Qt and AI/ML-related libraries for Fedora Linux helping bring cutting-edge open source tools to millions of users through the official Fedora repositories. I am also an Organizer and Speaker at GDG Samsun, where I share insights on Deep Learning, Machine Learning, Computer Vision, and Vision-Language Models (VLMs).

In addition to my community work, I contribute to a variety of FLOSS and OSS projects across GitHub, GitLab, and Pagure always aiming to build together and make open source a better place for everyone.

📦 `rpm --build-pipeline`

| %prep | %build | %install | %check | dnf install |
|:---:|:---:|:---:|:---:|:---:|
| 🔵⚪⚪⚪⚪ | 🔵🔵⚪⚪⚪ | 🔵🔵🔵⚪⚪ | 🔵🔵🔵🔵⚪ | 🔵🔵🔵🔵🔵 |
| *Patch & unpack sources* | *Compile & link* | *Stage files* | *Run tests* | *Ship to users!* |

console
$ fedpkg clone python-awesome-package &amp;&amp; cd python-awesome-package
$ fedpkg prep       # ✅ Sources verified, patches applied
$ fedpkg build      # ✅ RPM built successfully
$ fedpkg mockbuild  # ✅ Mock build passed (x86_64, aarch64)
$ fedpkg push &amp;&amp; fedpkg submit  # 🚀 Submitted to Bodhi
&gt; Update pushed to stable → reaching ~40M Fedora users

⚡ `cuda_oxide::kernel::launch()`

Currently exploring cuda-oxide an experimental Rust-to-CUDA compiler from NVIDIA Research that compiles safe(ish), idiomatic Rust directly to PTX. No CUDA C, no DSLs, no foreign bindings — just Rust running on the GPU.

| Grid | Block 0 | Block 1 | Block 2 | Block 3 |
|:---:|:---:|:---:|:---:|:---:|
| Warp 0..3 | 🟩🟩🟩🟩🟩🟩🟩🟩 | 🟦🟦🟦🟦🟦🟦🟦🟦 | 🟨🟨🟨🟨🟨🟨🟨🟨 | 🟥🟥🟥🟥🟥🟥🟥🟥 |
| SIMT | *all threads run the same instruction* | ← | ← | ← |

rust
// 🦀 Safe(ish) GPU kernel — Rust compiles straight to PTX
#[kernel]
pub unsafe fn vector_add(a: &amp;[f32], b: &amp;[f32], c: &amp;mut [f32]) {
    let tid = thread::index_1d() as usize;
    if tid &lt; c.len() {
        c[tid] = a[tid] + b[tid]; // thousands of threads, one instruction
    }
}

console
$ cargo build --target nvptx64-nvidia-cuda
   Compiling cuda-kernel v0.1.0
   ✅ Rust  -&gt; LLVM IR  -&gt;  PTX   (no CUDA C required)
   ✅ Kernel compiled: vector_add.ptx
   ✅ Loaded onto device: NVIDIA GPU
&gt; Launching 4096 threads across 128 warps...
&gt; All threads converged. ✅
&gt; Status: still learning — but the 🦀 loves the GPU

🔧 `deploy::stack()`

python
stack = {
    &quot;backend&quot;:   [&quot;Python&quot;, &quot;FastAPI&quot;, &quot;Django&quot;, &quot;Flask&quot;],
    &quot;mobile&quot;:    [&quot;Kotlin&quot;, &quot;Flutter&quot;, &quot;ARCore&quot;, &quot;MediaPipe&quot;, &quot;Firebase&quot;],
    &quot;inference&quot;: [&quot;ONNX Runtime&quot;, &quot;TensorRT&quot;, &quot;ExecuTorch&quot;, &quot;ncnn&quot;, &quot;Vulkan&quot;],
    &quot;training&quot;:  [&quot;PyTorch&quot;, &quot;Ultralytics&quot;, &quot;TensorFlow&quot;, &quot;JAX&quot;, &quot;Keras&quot;],
    &quot;gpu&quot;:       [&quot;CUDA&quot;, &quot;cuda-oxide&quot;, &quot;PTX&quot;, &quot;NVIDIA TensorRT&quot;],
}

On the backend, I primarily use Python, while on the mobile side, I work with Kotlin and Flutter, leveraging Google technologies such as ARCore, MediaPipe, Firebase, FlatBuffers, ONNX Runtime, and ncnn.

On the ML side, I work extensively with Ultralytics, where I am a Senior Machine Learning Engineer, developing and maintaining cutting-edge computer vision tools and YOLO models. My primary deep learning framework is PyTorch, which I use for model training, fine-tuning, and deployment workflows.

I also contribute to related open-source projects such as ExecuTorch an efficient on-device inference framework from PyTorch, and ONNXSlim a lightweight ONNX optimization toolkit.

In addition, I have experience with other frameworks like TensorFlow, TensorBoard, Keras, and JAX, which I occasionally use for experimentation and cross-framework integration.

🌍 `oss::contributions()`

rust
let platforms = vec![&quot;GitHub&quot;, &quot;GitLab&quot;, &quot;Pagure&quot;];
let mission   = &quot;build together — make open source better for everyone 🌍&quot;;

In addition to my diverse skill set, I am actively contributing and co-maintaner to a computer vision project named sahi is an open-source project that provides a simple and efficient way to perform slicing-based inference for object detection models. It is designed to work with various deep learning frameworks and models, making it a versatile tool for computer vision tasks. In past I was a co-maintaner to Supervision. Supervision is dedicated to crafting reusable computer vision tools tailored to a wide array of needs. These projects are embodies my commitment to advancing the field of computer vision, offering robust solutions that empower developers and researchers alike.

I am also contributing various FLOSS and OSS projects around GitHub/Gitlab/Pagure and other platforms to make everyone happy and doing better projects together.

🌱 Hacktoberfest Journey

I’ve proudly participated in Hacktoberfest every year since 2022, completing four consecutive years (2022, 2023, 2024, 2025).
Through this journey, I’ve contributed dozens of PRs, met inspiring developers from around the world, and grown both technically and personally.

💚 For me, Hacktoberfest isn’t just about code, it’s about people, learning, and giving back to the community. I’m also part of the Holopin x Hacktoberfest: 10 Badge Club, celebrating those who’ve contributed continuously since 2022. And yes, somewhere out there, a tree is growing in my name — a small but meaningful reminder that code can make a difference beyond the screen 🌍.

🧩 Let’s Connect

I’m always open to collaborating, contributing, or just talking about open source, computer vision, and AI.
Feel free to reach out or check out my latest work here on GitHub! 🚀

Here is my Arsenal Tools and Skills I use;

Programming

Backend-as-a-Service (Baas)

Web Servers

Framework & Libraries

<img alt="PyTorch" src="https://img.shie

…

All public repos (200)

Show forks Show archived

Onuralp SEZER

🧠 model.summary()

🚀 inference_pipeline.run()

🧬 model.fit(epochs=1000)