Data Scientist & AI Solutions Engineer specializing in LLMs, autonomous agents, and production-grade machine learning systems. Turning complex data into actionable intelligence.
Design and deploy autonomous AI agents for workflow automation, enterprise decision-support, and complex multi-step reasoning tasks.
End-to-end development of intent classification, ranking systems, and sentiment analysis using BERT, RoBERTa, and ensemble architectures.
Production-grade Retrieval-Augmented Generation systems. From vector databases to knowledge graphs, building scalable solutions for enterprise environments.
Statistical modeling, forecasting, and advanced analytics for business decision-making. From customer churn prediction to financial time-series analysis.
Topology-aware predictive models for complex network analysis, resilience assessment, and vulnerability detection in critical systems.
Digital banking assistants, chatbot optimization, and customer retention analysis through advanced NLP techniques.
Developing autonomous AI agents powered by LLMs for workflow automation. Designing secure on-premise AI architectures and integrating systems with REST APIs and unstructured data sources.
Led NLP and ML pipeline enhancements for digital banking assistants at Itaú Bank. Built classification and ranking systems processing 1M+ customer interactions monthly using BERT, XGBoost, and ensemble architectures.
Applied research in machine learning and complex networks for Smart Grid resilience. Developed Graph Neural Networks for power system vulnerability analysis and implemented optimization algorithms for critical infrastructure.
Itaú Bank — NTT DATA
Led the complete overhaul of NLP pipelines for one of Latin America's largest digital banking assistants. The system processes over 1 million customer interactions monthly, handling intent classification, sentiment analysis, and escalation risk detection in real-time.
Specialized in adapting large language models (GPT, BERT, T5) for enterprise use cases through fine-tuning with domain-specific datasets. This includes experimental comparison of training architectures, preprocessing techniques optimization, and preparation of large-scale textual datasets. Developed custom evaluation frameworks to measure model performance on specific business metrics rather than generic benchmarks.
Architected and deployed autonomous AI agents powered by Large Language Models for complex workflow automation. These agents handle multi-step reasoning tasks, integrate with heterogeneous REST APIs, and process unstructured data from multiple sources. Built secure on-premise deployment pipelines for enterprise clients with strict data governance requirements, ensuring LGPD compliance throughout.
Conducted applied research at LabTel/UFES developing Graph Neural Networks to model power system resilience and vulnerability propagation in Brazil's energy transmission network. Created topology-aware predictive models for failure detection and ranking, enabling proactive maintenance strategies. Applied Differential Evolution optimization algorithms to improve transmission line risk assessments for critical infrastructure protection.
Developed a comprehensive customer risk scoring system for digital banking that identifies high-risk interactions before they escalate to human agents. Built ranking algorithms combining NLP-based sentiment analysis with behavioral patterns to prioritize interventions. The recommendation mechanism increased digital service completion rates by surfacing personalized solutions at critical conversation moments.
Built a deep learning model using LSTM networks to forecast NVIDIA stock price movements based on 8 years of historical data (2016-2024). Implemented feature importance ranking and sequence modeling techniques to identify the most predictive signals. The project explored time-series ranking algorithms to determine which historical patterns have the strongest predictive power for future price action.
Conducting econometric analysis of how central bank interest rate decisions (Brazil's Selic and US Fed Funds Rate) affect Brazilian industrial stock performance on B3. Developing ranking models and performance scoring algorithms to identify which sectors show the strongest correlation with monetary policy shifts, providing quantitative insights for risk assessment and investment strategy formulation.
Applied clustering and analytical modeling to understand employee absence patterns, with a focus on mental health leave analysis. Used unsupervised learning techniques (K-Means, DBSCAN, hierarchical clustering) to identify risk segments and inform HR policy decisions. The insights helped optimize resource allocation and develop proactive support programs for at-risk employee groups.