Computational Physicist → ML/Data Science for Materials
Postdoctoral Researcher @ Uppsala University · PhD in Physics (USP) · MBA in Data Science (ESALQ/USP)

I build end-to-end pipelines that turn physics simulations into predictive ML models — from raw HPC outputs to published datasets and production-ready tools.

• ML for Materials Science — Graph Neural Networks, Random Forest and Gradient Boosting to predict structural and electronic properties of materials
• High-throughput workflows — automated pipelines for large-scale first-principles simulations on HPC clusters (SLURM, Python, YAML-driven)
• Data engineering — from simulation output → structured dataset → publication. Built and released TBHubbard, a 10,000+ MOF database published in Nature Scientific Data
• Predictive modelling — correcting systematic DFT errors in formation enthalpies using ML (active research, Uppsala University)
• MLOps / production ML — packaging research models into APIs, dashboards, Docker, CI, and AWS-hosted demos

Python toolkit for automating high-throughput EMTO physics simulations on HPC clusters

• Single YAML config drives input generation, parameter sweeps, SLURM scheduling and results extraction
• Designed for reproducibility and scalability across hundreds of concurrent jobs
• Python · pandas · scipy · matplotlib · SLURM

Predicting Hubbard U and V parameters in MOFs using the dataset I built and published

• Benchmarked Linear Regression and Random Forest against DFT reference values on 464,509+ pair-level observations from 242 MOFs
• Best model for U: Linear Regression (R² = 0.970, MAE = 0.181 eV); best model for V: Random Forest (R² = 0.782, MAE = 0.026 eV)
• Full pipeline: EDA → preprocessing (IQR outlier filtering, VIF multicollinearity check, group-aware train/test split) → modelling → evaluation
• Based on the TBHubbard dataset — 785+ downloads since release
• scikit-learn · statsmodels · pandas · matplotlib · seaborn

From thesis notebooks to a deployable ML system: train → register → serve → demo

• Serves U/V Hubbard predictions with MLflow model registry (@champion), FastAPI, and a Streamlit dashboard (Explore / Predict live / Prediction log)
• Containerized with Docker Compose and deployed on AWS EC2; GitHub Actions CI (train → pytest → image build)
• Live demo: dashboard · walkthrough video
• Built on top of TBHubbard ML Analysis
• scikit-learn · MLflow · FastAPI · Streamlit · Docker · AWS EC2 · pytest · GitHub Actions

• 🥇 Best Physics Phd Thesis in 2025 — awarded by the Physics Institute in the University of São Paulo
• 🥇 Bernard Gross Prize — Best oral presentation, XXI B-MRS Meeting (Brazilian Materials Research Society, 2023)
• 🔬 Research internships at IBM Research Brazil and Uppsala University during PhD