Darlan Barroso

I develop geospatial intelligence platforms using Machine Learning models (Scikit-learn, CatBoost) for market viability analysis and business opportunity identification, integrating data from sources such as Google Places, Google Trends, and business registries, with interactive heatmap visualizations. I design and implement multi-agent systems to orchestrate intelligent conversational assistants using Google ADK, LangGraph, and the MCP protocol, applied to use cases such as customer support, course recommendations, and educational support with RAG (Retrieval-Augmented Generation). I build generative AI solutions (Gemini / Vertex AI) for personalized learning, transforming educational content into multiple formats — audio, transcriptions, summaries, and quizzes — integrated into learning platforms and compliant with data privacy regulations. I work on the architecture and development of cloud-native applications on GCP, using FastAPI, React/Next.js, PostgreSQL, Redis, and Docker, including deployment of conversational agents on Vertex AI Agent Engine and ML/RAG pipelines on GKE, with observability through OpenTelemetry and Prometheus.

ArcelorMittal is a global leader in steel and mining, recognized for producing a wide range of high-quality steel products and for pioneering sustainable solutions in steel production. - I am developing a computer vision system using YOLOv8 to detect people within the operational range of overhead cranes, integrated with an audio-visual alarm interlock system to enhance workplace safety. - I work on reverse engineering initiatives to improve machines and industrial processes, increasing efficiency and cost savings, enhancing production reliability, and reducing downtime.

LABIOTEC brings together researchers to analyze, develop, and optimize industrial biotechnological products and processes, including applications in the food industry. - Increased data extraction efficiency by 80% and reduced manual effort by developing an automated script using Selenium and BeautifulSoup to collect scientific articles from the web. - Reduced analysis time by 40% and improved data organization and accessibility by converting HTML tables into structured Excel datasets, using article titles as the primary reference column.

Photonis Lab focuses on research in integrated optics, collaborating on R&D projects with industry partners and funding institutions to enhance the performance of sensors and optical systems. - Achieved 80% prediction accuracy by developing an API with a machine learning backend to estimate photonic properties of nanostructures. - Achieved 90% prediction accuracy by developing an API to predict properties of optical communication systems. - Conducted research on nanostructure properties, advancing understanding and exploring potential applications.

Aeris Energy is a leading manufacturer of wind turbine blades, committed to advancing renewable energy and sustainability through innovative design and precision engineering. - Reduced inspection time by 75% (from 2 hours to 30 minutes) and decreased measurement errors by 30% by developing a computer vision system for automatic paint thickness measurement using Python and OpenCV. - Supported the Quality Management System through data analysis, process control, audits, and training, performing complex activities requiring strong technical expertise. - Developed photogrammetry measurement templates using Python scripts, improving accuracy and operational efficiency. - Conducted statistical analyses for Measurement System Analysis (MSA) in blade lamination and finishing processes, ensuring accuracy in quality control.

- Área de Pesquisa: Design direto e inverso de nanopartículas de ouro, prata, bronze e silício. - Ferramentas Utilizadas: - PyCaret: Para seleção dos melhores modelos de machine learning. - COMSOL: Para obtenção de dados via simulação computacional. - Modelos de Machine Learning Selecionados: - Random Forest: Implementado com scikit-learn. - Gradient Boosting: Implementado com xgboost. - Extra Tree: Implementado com scikit-learn. - Desenvolvimento de API: - Utilização de Streamlit para criação de uma API para predição de dados desconhecidos usando o melhor modelo de machine learning. - Futuras Direções: - Emprego de técnicas de Deep Learning para realizar o design inverso. - Aplicação de modelos avançados a dados experimentais para aprimorar precisão e eficiência.

Desenvolvimento (Python) de um sistema de visão computacional utilizando processamento digital de imagens para medição automática da espessura do revestimento de tinta aplicado em pás de turbinas eólicas.