Arkadii M.

I am a highly skilled AI/ML Engineer specializing in building intelligent, production-grade AI systems. My expertise spans Artificial Intelligence, Machine Learning, Natural Language Processing (NLP), Deep Learning, Computer Vision, and AI Agent Development. I am proficient in Python and work extensively with frameworks such as LangChain, LangGraph, TensorFlow, PyTorch, and Hugging Face Transformers to design, fine-tune, and deploy complex models. I have hands-on experience building autonomous AI agents and LLM-powered applications, including a stateful conversational assistant (LangGraph, OpenAI API) for a food delivery and wellness app with personalized recommendations, nutrition tracking, and image-based food recognition, and an autonomous media-buying agent powered by the Anthropic Claude API that integrates with the Meta Marketing API for automated budget optimization with human-in-the-loop approval, real-time SSE streaming, and voice command processing (Deepgram). In fraud detection and anomaly analysis, I've built and maintained a large-scale claims intelligence platform (~7.8M+ records) combining deterministic business rules, behavioral clustering, and an ensemble ML pipeline (Isolation Forest, One-Class SVM, LOF) with SHAP explainability, backed by FastAPI, SQL Server, and SQLAlchemy. My Computer Vision experience includes fine-tuning YOLOv8 for object detection on satellite imagery (parking space detection) using tiling-based training for improved accuracy, along with building Streamlit demos for interactive model evaluation. I also have experience with Retrieval-Augmented Generation (RAG) for document analysis (legal document summarization and Q&A using LangChain and vector databases like Pinecone/FAISS), text classification and AI-generated content detection using fine-tuned Transformer models (RoBERTa, DeBERTa) combined with statistical NLP methods, and building customer support automation systems using Dify, BERT, and Elasticsearch-based RAG pipelines. I work closely with data to drive impactful, measurable results, and I'm proficient in API integrations, cloud deployment (AWS, Azure), Docker-based infrastructure, and observability/evaluation tooling (Langfuse, Weights & Biases) to ensure AI systems are reliable, explainable, and production-ready. My goal is to deliver innovative, high-performance AI solutions — from LLM agents to fraud detection systems and computer vision pipelines — that solve real-world problems and drive measurable business value.