Professional Summary
Senior AI engineer with 6+ years shipping production-grade LLM, generative-AI and
computer-vision systems. I design scalable AI/ML platforms — multi-agent LLM
systems, RAG and retrieval/ranking pipelines, LLM evaluation, and hybrid setups
that pair generative models with deterministic processing. Proven, measurable
impact: ~75% faster training, +15pp IoU over commercial baselines, and pipelines
serving 300K+ samples (up to 1M points). Strong hands-on background in PyTorch,
distributed training, inference optimization, MLOps and cloud, with end-to-end
delivery alongside domain experts and stakeholders.
Technical Skills
- LLM & GenAI
- Multi-agent systems, RAG, HITL, MCP, LangChain, LangGraph, PydanticAI, LangFuse, Hugging Face
- Generative & Foundation Models
- Diffusion, Latent Diffusion, Rectified Flow, GANs, VAEs, multimodal models
- Computer Vision & 3D
- YOLO, OpenCV, PyTorch3D, Open3D, Trimesh, NeRF / NeuS, mesh & point-cloud processing
- Training & Optimization
- DeepSpeed ZeRO, multi-GPU, bf16 / FP16 mixed precision, FlashAttention, LoRA / PEFT, quantization
- ML Frameworks
- PyTorch, PyTorch Lightning, TensorFlow, scikit-learn, XGBoost
- Backend & APIs
- FastAPI (async), REST, pytest / pytest-asyncio, Locust, OpenTelemetry, structlog
- MLOps & Cloud
- Docker, Kubernetes, CI/CD, MLflow, W&B, GCP, AWS (S3, EC2), Huawei Cloud (ModelArts), Git, Linux
- Data & Storage
- PostgreSQL, asyncpg, Redis, Elasticsearch, PySpark, Pandas, NumPy, FAISS, pgvector, SQL
Professional Experience
DentalTwin — Senior AI Engineer / ML Systems Architect
Jan 2024 – Feb 2026 · Germany (Remote)
3D Generative AI · LLM Systems · MLOps · Production deployment.
- Designed and productionized large-scale 3D generative AI (Diffusion, Rectified Flow, GANs, VAEs) on 300K–350K+ samples (up to 1M vertices) — modular framework, unit-tested, Dockerized, K8s-ready.
- Cut training time ~75% (12 → 3 days) with DeepSpeed ZeRO, bf16 mixed precision, FlashAttention and LoRA/PEFT across cost-efficient multi-cloud GPUs.
- Built LLM evaluation pipelines (LangChain + LLM APIs) and exposed geometry-processing libraries as MCP tools for LLM-driven, tool-augmented workflows.
- Ran human-in-the-loop feedback with domain experts via a custom A/B + rating annotation tool; collected preference data, retrained models, and tracked win-rate to catch quality regressions.
- Improved 3D segmentation mean IoU ~15pp (≈80% → ≈95%), surpassing commercial baselines for manufacturing-grade applications.
- Built scalable async FastAPI inference (multi-user, monitored, Locust-load-tested) and large-scale 3D data pipelines with multi-GPU CI/CD (MLflow, W&B).
Tools: PyTorch, Diffusion/Rectified Flow, DeepSpeed, LangChain, MCP, FastAPI, Docker, K8s, GCP, MLflow, W&B
Huawei (R&D Division) — AI Engineer
2022 – 2024 · Türkiye / Global
Multimodal AI · NLP/LLM · Computer Vision · Search & Retrieval.
- Core contributor in a 500+ engineer global R&D initiative, building multimodal AI systems and PoCs across NLP, vision and time-series, moving research into production.
- Designed LLM-driven retrieval & ranking (RAG) combining layout detection (YOLO), semantic embeddings, dense retrieval, NER and FAISS vector search for automated resume-job matching.
- Built a global-scale sign-language production system (transformer LLMs, Text→Sign, Sign→Pose) with aligned text/3D-pose pipelines on Huawei Cloud.
- Developed LLM augmentation via function-calling APIs for synthetic training data and structured outputs; optimized transformer/vision models for Ascend hardware (MindSpore).
- Benchmarked time-series forecasting (FEDformer, Autoformer, Informer) on multi-source telecom data; built energy-consumption forecasting PoCs (LSTM, Transformer, XGBoost).
Tools: PyTorch, Transformers, RAG, FAISS, YOLO, MindSpore, Huawei Cloud
NTT DATA — Data Scientist
2021 – 2022 · Türkiye
- Designed and deployed scalable ML for retail optimization; improved object-detection mean IoU 25% → 65% (+40pp).
- Collaborated with data engineering on cloud-deployed ML components and ran AI training sessions to accelerate ML adoption across business units.
Tools: PyTorch, OpenCV, cloud ML
PCIS Consultancy — Software Engineer
2020 – 2021 · Türkiye
- Built real-time industrial IoT systems (.NET, JavaScript) for production monitoring and SAP-integrated data pipelines.
- Developed early computer-vision modules for automated inspection workflows.
Selected Projects
- Conversational AI backend (automotive client) — qualifies leads and captures requests over WhatsApp & web; multi-agent orchestration (FastAPI, PydanticAI) with ~10–12 CRM tools, multi-turn history on PostgreSQL, LangFuse observability, Redis caching and JWT auth.
- Agent-based LLM analysis system — structured content analysis with LangChain: RAG, tool use and modular agent orchestration.
- Open-source medical image segmentation (U-Net) — widely adopted segmentation pipeline.
- End-to-end OCR pipeline — image preprocessing → segmentation → structured text extraction.
Education
MSc in Data Science — Yeditepe University · 2021–2024
BSc in Biomedical Engineering — Yeditepe University · 2015–2020
Author of 6+ peer-reviewed AI/ML publications (IEEE, arXiv).