Milan, Lombardy, Italy
Vertical Technical Leader specializing in agentic AI systems powered by vision-language models (VLMs). I design and deploy production pipelines combining VLMs, LLMs, OCR, and layout models for unstructured information extraction from documents, including visual and tabular data. Currently leading AI product development at Orobix, architecting agentic systems for intelligent document search, retrieval, and information extraction from complex technical documents. Previously built and shipped deep learning systems for computer vision across pharmaceutical, manufacturing, and agricultural domains: anomaly detection, segmentation, object detection, and quality analysis. Core expertise: - Agentic AI & Agent Engineering (LangGraph, PydanticAI, MCP) - Vision-Language Models (VLMs) & Large Language Models (LLMs) - Document Understanding & Information Extraction (OCR, tables, layout extraction) - Computer Vision (anomaly detection, segmentation, object detection) - MLOps & Production ML (PyTorch, FastAPI, Docker, Azure)
Led development of Generative AI solutions combining VLMs, LLMs, vision, OCR, and layout models for unstructured information extraction from documents, including visual and tabular data, powering advanced PDF understanding for Teki Product. Designed and deployed agentic workflows using LangGraph and FastAPI, including custom agent and tool design and MCP server integration for scalable, modular orchestration. Technologies: Python, PyTorch, LangGraph, PydanticAI, FastAPI, Azure, VLMs, LLMs, OCR, MCP
Led the Quadra open-source project, a reproducible deep learning experiment management framework integrating PyTorch Lightning and Hydra for scalable ML workflows. Built and optimized deep learning models (classification, segmentation, anomaly detection) for the pharmaceutical and packaging sectors, powering the Aigo visual inspection platform. Applied computer vision and deep learning for plant disease and fruit quality analysis, improving yield quality and efficiency across on-field and post-harvest processes for QUALYFruit. Developed reinforcement learning agents in PyTorch for simulations and games, achieving improved decision efficiency and seamless cross-platform integration. Technologies: Python, PyTorch, PyTorch Lightning, Hydra, Computer Vision, Deep Learning, Reinforcement Learning
Designed and trained meta-learning algorithms using distributed on-premise GPU infrastructure with PyTorch, improving training efficiency and scalability. Built efficient project tracking and experiment configuration tools to streamline AI model training workflows. Technologies: Python, PyTorch, Distributed Computing, Meta-Learning
- Designed face recognition, age, and gender estimation software from face images - Developed edge computing mobile applications powered by deep learning and computer vision. - Practiced with Tensorflow and Keras libraries
- Implemented RESTful client-side of an Android application. - Optimized server connection speed and performance of `Turkcell Hesabım` Application.
- Designed and simulated a high frequency & high gain satellite space antenna with HFSS tool. - Worked on horn antennas and waveguides.