Physicist (UFSCar) with an MBA in Artificial Intelligence and Big Data (USP).
I apply data engineering, machine learning and physics-informed models to aircraft reliability, predictive maintenance and engine prognostics.

• ✈️ Engineering Analyst — Reliability Engineering at GOL Linhas Aéreas: building predictive models, failure-pattern and root-cause analyses, and dashboards that improve the reliability of the Boeing 737 fleet
• 🎓 MBA in Artificial Intelligence and Big Data — Universidade de São Paulo (USP ICMC), completed 2026. Monograph: Applying Neuroevolution to the Architecture Search of Physics-Informed Neural Networks, advised by Prof. Dr. Rodrigo Colnago Contreras
• 🎓 B.Sc. in Physics — Universidade Federal de São Carlos (UFSCar), 2025, with undergraduate research in applied acoustics
• 🔭 Focused on reliability engineering and prognostics (RUL, Weibull, MTBF), physics-informed ML, and end-to-end data pipelines
• 🌱 Next on the roadmap: security data engineering and detection engineering, bringing the same data-driven approach to cybersecurity
• 💬 Ask me about Python, SQL, Spark, PINNs, reliability metrics, or predictive maintenance
• 🌎 Based in Mogi Guaçu, SP, Brazil

• Aviation reliability: Weibull failure modeling, MTBF / MTTR / availability, failure Pareto analysis, star-schema reliability data models and interactive fleet dashboards
• Prognostics & physics-informed ML: Remaining Useful Life prediction on NASA N-CMAPSS, PINNs with physics-based loss terms, neuroevolution and NSGA-II architecture search, fault detection from vibration and engine sensor data
• Optimization: genetic algorithms (DEAP), multi-objective NSGA-II Pareto fronts, and exact CP-SAT models (OR-Tools) for fleet maintenance scheduling under hangar, labour and spare-part limits
• Aerodynamics from first principles: thin airfoil theory, panel methods, integral boundary layer and vortex lattice method, each validated against exact solutions and wind-tunnel data
• Data engineering: ELT with dbt star schemas, Data Lake ↔ Data Warehouse integration, data-quality handling and quarantine, batch (Airflow) vs event-driven serverless (AWS + Terraform) architectures, distributed Spark clusters
• ML engineering: experiment tracking (MLflow), model serving with FastAPI, Streamlit dashboards, Docker Compose stacks, automated tests and CI on every project
• Rigorous evaluation: detecting and documenting data leakage, unit-level train/test splits, multi-seed results, and honest baselines (classical models are hard to beat)
• Applied AI: CNN transfer learning with Grad-CAM explainability, NLP on aviation safety reports, LLM apps grounded with RAG (Qdrant), time-series forecasting with Monte Carlo risk (VaR/CVaR)
• Security data: network intrusion detection on NSL-KDD and CICIDS2017 with Spark MLlib
• Simulation & systems programming: C++20 ecosystem simulator with ECS, multithreading, sanitizers and profiling-driven optimization

Languages & Data

Data Engineering

Machine Learning & AI

Optimization & Reliability

Cloud, DevOps, BI & Tools

🔗 Remaining Useful Life Prediction with PINN and Neuroevolution Predicts turbofan engine RUL on NASA's N-CMAPSS dataset with a Physics-Informed Neural Network whose architecture and physics-loss weights are tuned by an NSGA-II neuroevolutionary search, compared against MLP, 1D-CNN and LSTM/GRU baselines. PyTorch, pymoo.

🔗 Preventive Maintenance Optimization with Genetic Algorithms Schedules a year of preventive maintenance for a 10-aircraft fleet under hangar, labour and spare-part limits. The GA cuts expected cost by ~25% versus cost-optimal fixed intervals and lands within ~1% of an exact CP-SAT model; NSGA-II maps the cost-vs-risk trade-off. DEAP, OR-Tools.

🔗 Aircraft Fleet Reliability Dashboard Weibull-based synthetic maintenance history → PostgreSQL star schema → validated SQL metrics layer → Streamlit dashboard with MTBF, MTTR, availability, failure Pareto, Weibull hazard analysis and downtime cost.

🔗 Fault Detection in Engine Parts (Physics-Informed Deep Learning) Fault diagnosis for bearings and turbofan parts: classical baselines, 1D-CNN, autoencoder, graph neural network, Optuna/PSO tuning and a physics-informed defect-frequency loss, served through an API and dashboard.

🔗 Aerodynamics From Scratch Thin airfoil theory, a 2D panel method, an integral boundary layer and a 3D vortex lattice method, written from first principles in NumPy — every solver validated against exact solutions or published wind-tunnel data.

🔗 Neuroevolution for the Architecture Search of PINNs — MBA Monograph (USP) A cheap single-generation Genetic Algorithm that selects PINN architectures from short partial trainings, evaluated on the pendulum ODE, heat and wave equations over 35 seeded runs per problem against random choice and random search.

🔗 Data Warehouse ↔ Data Lake Integration with ELT Messy raw data → cleaned lake files (Polars, PyArrow, DuckDB) → raw landing zone → dbt-modeled star schema in PostgreSQL, with data-quality quarantine, tests and generated lineage. Fully containerized.

🔗 Currency Rates — Two Architectures: Airflow vs AWS Serverless The same exchange-rate pipeline built twice: self-hosted batch with Airflow + PostgreSQL, and event-driven on AWS (Lambda, S3, EventBridge, Glue, Athena) defined in Terraform within the free tier.

🔗 Fraud Detection with Spark ~1.5M synthetic transactions from three messy source systems cleaned on a real Spark cluster, a GBT classifier trained with MLlib, live scoring through FastAPI and a Streamlit dashboard — all on Docker Compose with CI.

🔗 Network Intrusion Detection NSL-KDD and CICIDS2017 processed on a Spark cluster, independent MLlib classifiers per benchmark, served with FastAPI and visualized in Streamlit.

🔗 Brain Tumor MRI Classifier Four comparable CNN architectures (baseline + MobileNetV2, EfficientNetB0, ResNet50) with MLflow tracking, a held-out test set, Grad-CAM explainability, a REST API and a web UI.

🔗 Smart Pokédex — Django + LLM + RAG A Django app consuming the PokéAPI where users ask an LLM about each Pokémon, with answers grounded by RAG over a Qdrant vector database.

🔗 Cash Flow Forecasting with Time Series and Monte Carlo SARIMA and Prophet forecasts with a Monte Carlo layer for confidence bands and risk scenarios (VaR/CVaR).

🔗 Predicting and Analyzing Diabetes A rebuild of an earlier project that claimed 100% accuracy — with a documented investigation of the data leakage behind that number.

🔗 Ecosystem / Wildlife Simulator A C++20 simulation where predator–prey cycles, epidemics, seasonal migration and competitive exclusion emerge from local rules alone — validated against exact results, profiled, multithreaded and tested with sanitizers.

📫 Reach me at [email protected], on LinkedIn or through my portfolio