Özgün Deniz Beyaz

I'm an AI Engineer with 2+ years of shipping generative AI into daily operations: RAG, text-to-SQL and multi-agent systems.

I build LLM systems that have to hold up on a Monday morning, not just in a demo. Most of my work is Python: FastAPI services, with LangChain for orchestration. I choose models per task, not per brand: GPT, Gemini, Claude, Llama or DeepSeek, depending on the latency, cost and quality the job needs.

Most of what makes an agent dependable is not the model but the harness around it, so that is where I put most of my effort: harness engineering. The tools an agent can call, the context it sees at each step, the checks its output has to pass, and the point where it hands the decision back to a person.

I came to AI from Geomatics Engineering at Istanbul Technical University. The work there was remote sensing and measurement: noisy signals, imperfect sensors, and the habit of never trusting a number you haven't validated. I carried that habit into AI, where evaluation is part of the system rather than an afterthought. I later specialized in machine learning and deep learning through DeepLearning.AI.