Ugur Demirkaya

AI Engineer specializing in agentic systems that turn messy enterprise data — transaction logs, scanned documents, decades of unstructured archives — into automated decision infrastructure. I design multi-agent architectures and RAG pipelines (LangGraph, GraphRAG, FAISS, ChromaDB) with local and cloud LLM inference, shipped as production-ready microservices using FastAPI, Docker, and MLOps tooling (MLflow, ZenML).

Recent projects include a 5-agent fraud detection platform reaching AUC-ROC 0.919 with full Turkish-language explainability, and an executive RAG chatbot ingesting 50+ years of mixed-format meeting archives (OCR, DOCX, scanned PDFs) that cut retrieval time by 50%. I also bring classical ML (XGBoost, LightGBM) and OR-based optimization (Google OR-Tools, MILP) into projects where forecasting or resource scheduling is involved.

I work end-to-end — from data pipeline design to model development to production deployment — and I'm available for remote freelance projects, whether that's building a RAG system from scratch, adding an agentic layer to an existing pipeline, or optimizing an ML system already in production. Open to short-term projects and ongoing collaborations.