Tuğrul A.

I build machine learning systems that work outside the notebook. A planner that now runs a live five-lift car park, a camera model built to replace €2,000 laser scanners (98.9% validation accuracy), real-time depth estimation at 60 fps on TensorRT. What I love most is making machines learn and integrating their capabilities into human workflows. I work across machine learning and AI, including model development and training, edge AI deployment, computer vision, and data analysis. I'm comfortable taking a project from the data stage through to something running on real hardware. On the LLM side, I build tools and agents with the guardrails in code, not in the prompt. I also do API integration and systems work, getting systems to talk to each other that were never designed to. Undocumented platforms, mismatched formats, tools from completely different worlds that have to work together. I've been doing that for about a decade. When a model needs a person in the loop, I build that part too. Native desktop applications in GTK and Qt, internal tools, custom interfaces for hardware and data work. This is where a lot of ML projects stall, and it's ordinary work for me. What you get at the end is a repo that runs on your machine and not just mine. Clear structure, setup steps that actually work, and the reasoning behind the non-obvious decisions written down. A lot of ML work gets handed over as a notebook only its author can run. I'd rather you be able to extend it later, with or without me. My usual stack is Python (PyTorch, scikit-learn, OpenCV, NumPy, pandas), TensorRT and ONNX, LLM APIs (Anthropic, OpenAI, Gemini, local models with llama.cpp), C++, GTK, Qt, Linux, Flask/FastAPI, React/TypeScript, plus SQL and NoSQL databases.