At N2Mobil, I build production AI systems and backend services, including Niki, an AI-powered IoT fleet diagnostics platform. I architected its backend and multi-agent system with LangGraph and built a Qdrant-based Hybrid RAG pipeline to ground responses in technical documentation.
I built a vLLM-based multi-LoRA inference server for Qwen3-VL-8B, serving 200+ daily requests on an RTX 5090 server. QLoRA quantization and Unsloth optimization reduced infrastructure cost by about 40% compared with running separate full models.
For Ruhsat Reader, I developed the Turkish vehicle registration OCR pipeline from dataset preparation and QLoRA fine-tuning through FastAPI deployment and testing. It reached 99%+ field-level accuracy after normalisation on a held-out test set of 500+ real-world documents.
I also deployed an on-premise Ollama code review pipeline at N2Mobil so proprietary source code stayed inside the internal network; initial review analysis time dropped from up to two days to about five minutes. Earlier, I worked on industrial computer vision at Büyütech and delivered full-stack, ML, and automation projects through Upwork & Freelance.

