Design and deploy practical AI systems (LLMs + Computer Vision) with a strong emphasis on reproducibility, performance, and end-to-end observability—turning research experimentation into maintainable production workflows.
| Category | Focus |
|---|---|
| Languages / Scripting | Python, C# (Unity) |
| Deep Learning | PyTorch, YOLO (v8 / v9 / v11) |
| Serving / APIs | FastAPI (inference & microservices) |
| Experiment & Lineage | ClearML, MLflow |
| Model Optimization | ONNX, NCNN (runtime benchmarking) |
| Computer Vision | Detection, classification, multi-attribute pipelines |
| Tooling & Infra | Docker (containerized reproducible setups) |
| Emerging | LLM finetuning & evaluation, AR + AI convergence |
| Project | What It Does | Stack | Notes |
|---|---|---|---|
| Food-classification | Multi-class food image classifier | PyTorch | Food-101 (~101k images); ~85% top-1; sub-30ms ONNX inference |
| Computer-Vision-MLOps-Toolkit | Reusable CV experiment & tracking toolkit | PyTorch, ClearML/MLflow, FastAPI | Standardized templates cut setup time by ~60% |
| Research-Learning-for-LLM-s | LLM experiments, notes & prototypes | Python, notebooks | Tasks: summarization & Q/A; Benchmarks: SQuAD, GSM8K |
| bulAR | AR note-taking concept for physical books/spaces | Unity, C# | Microsoft for Startups Founders Hub support |
- 40% inference speed improvement adopting NCNN over baseline PyTorch / ONNX for targeted models.
- Automated experiment tracking & versioning with ClearML (configs, dataset lineage, artifacts).
- Multi-format evaluation pipeline (PyTorch → ONNX → NCNN) guiding deployment runtime choices.
- Authored internal object detection labeling workflow docs (reduced onboarding friction).
- Built multi-attribute vehicle detection & classification (color / brand / model) with YOLO variants + OpenCV real-time.
- Containerized MLOps tooling for standardized scalable training environments.
- Founded bulAR (AR + note-taking) with Microsoft for Startups Founders Hub support.
| Pillar | Description | Why |
|---|---|---|
| LLM Experimentation | Finetuning, prompt patterns, eval loops | Faster domain adaptation |
| Runtime Optimization | Cross-format benchmarking | Lower latency & cost |
| CV Systems | Robust detection & classification | Real-world reliability |
| MLOps Foundations | Tracking, lineage, containers | Scalable, auditable workflows |
| AR + AI | Contextual note-taking interfaces | New interaction modalities |
Python • PyTorch • YOLOv8 / v9 / v11 • ClearML • MLflow • ONNX • NCNN • OpenCV • FastAPI • Docker • Unity (C#) • Streamlit • LLM (research & finetune) • MLOps
LLM finetuning & evaluation • ClearML / MLflow best practices • Computer vision detection pipelines • Runtime optimization (PyTorch → ONNX → NCNN) • Real-time inference & latency tuning • Reproducible ML experimentation • Bridging AR & AI
Looking for:
- Lightweight MLOps / tracking OSS
- Runtime benchmarking & optimization
- Real-time multi-attribute detection challenges
- AR + AI prototype concepts
I can help with:
- ClearML / MLflow setup & reproducibility
- Inference & evaluation loop design
- Model export & runtime benchmarking
- Structuring CV & LLM research repositories
| Dates | Role | Organization | Key Contributions |
|---|---|---|---|
| 06/2025 – 07/2025 | AI Engineer Intern | Mavinci | Experiment tracking, labeling docs, multi-runtime benchmarks, containerized stack |
| 01/2025 – 02/2025 | AI Engineer Intern | Huawei | ML/LLM study, preprocessing, training & optimization |
| 11/2024 – 01/2025 | AI Engineer Intern | Ucanble Technology | Vehicle detection & classification (YOLO), real-time CV |
| 07/2024 – 09/2024 | AI Engineer Intern | ArVis Technology | Energy systems DS project (Streamlit) |
| 12/2023 – 12/2024 | Founder | bulAR | AR note-taking concept, Unity prototype |
- B.Sc. Software Engineering, OSTİM Technical University (2022 – Present)
- Samsung Innovation Campus Program (2024 – Present)
- Quantization & efficient LLM serving
- Augmentation & robustness strategies in CV
- Unified metadata & lineage
- AR interaction paradigms + ML perception
- Reproducibility over ad-hoc speed
- Measure before optimizing
- Document as you build
- Containerize & version everything critical
- Benchmark with real latency / accuracy / cost signals
Email: [email protected]
LinkedIn: www.linkedin.com/in/seyit-ali-yorğun
Linktree: https://linktr.ee/Seyit_Ali
Core Lang: Python
DL: PyTorch | YOLO (v8/v9/v11)
Runtime Opt: ONNX | NCNN
Tracking: ClearML | MLflow
CV: Detection | Classification | Multi-attribute
Serving: FastAPI
Real-Time: OpenCV
Tooling: Docker | Streamlit | Unity (C#)
Domains: LLMs | Computer Vision | MLOps | AR+AI