Guilherme Hughes carcaraa

Guilherme de Sena Hughes

Computational Modeling (MSc) · ML & Scientific Computing · PINNs · PyTorch

I'm an electrical engineer turned computational scientist, currently pursuing a PhD in Computational Modeling at UESC (Brazil). My work sits at the intersection of physics-informed machine learning, numerical simulation, and data-driven operations. I like building things that connect mathematical rigor to real-world impact.

What I'm working on

🔬 Research - Physics-Informed Neural Networks (PINNs) for solving differential equations, coupled neutron-thermohydraulic simulations of nuclear reactor pins (Serpent + ANSYS Fluent), and multi-objective optimization with evolutionary algorithms.

⚙️ Industry - Leading data operations at scale (400-500 insurance certificates/week, zero-error SLA), building Python & Apps Script automation pipelines that eliminated hours of manual work, and developing demand forecasting models used for capacity planning.

Highlight: NeuTherm-PINN

End-to-end ML project solving the coupled neutron diffusion + heat conduction problem in a PWR fuel pin using PyTorch. Compares a data-driven surrogate (FNN) against a physics-informed neural network (PINN) that learns directly from the governing PDEs via automatic differentiation.

Component	Description
Solver	2-group finite-difference diffusion + Picard-coupled heat conduction ($k_{\text{eff}}$ = 1.30, converges in 6 iterations)
Surrogate	FNN with residual blocks trained on 5000 LHS samples — < 0.004% error on all fields
PINN	Learns $\phi_1(r), \phi_2(r), T(r)$ and $k_{\text{eff}}$ from PDEs alone, Learns φ₁(r), φ₂(r), T(r) from the PDEs + prescribed linear power; k_eff via integral neutron balance (1 reference solve, no field data)

Built from scratch: numerical solvers, cross-section models (Doppler feedback), dataset generation (Latin Hypercube), training pipelines, and comparison benchmarks. See the repo →

Featured projects

Project	What it does	Stack
neutherm-pinn-v2	PINNs for coupled neutronics–thermal hydraulics in nuclear fuel elements	PyTorch · NumPy · SciPy
redes-neurais	PINNs for solving ODEs/PDEs — full ML pipeline from linear regression to deep learning	PyTorch · NumPy · Matplotlib
article-multi-small-candu-th	Multiphysics simulation of a small supercritical CANDU reactor with thorium fuel	Serpent · ANSYS Fluent · Python
alocacao-dlccs	Multi-objective genetic algorithm (NSGA-II) for optimal placement of fault current limiters in power systems	Python · Optimization

Tech & tools

Languages        Python · C · R · JavaScript · SQL
ML / DS          PyTorch · Scikit-learn · Pandas · NumPy · SciPy · Matplotlib · Seaborn
Methods          PINNs · Regression · Classification · MLE · SGD · SVM · NSGA-II · Causal Inference
Simulation       ANSYS Fluent (CFD) · Serpent (Monte Carlo) · Numerical methods for ODEs/PDEs
DevOps           Docker · Git · Google Apps Script · ETL pipelines

Education

🎓 PhD in Computational Modeling - UESC (in progress)
📜 MSc in Computational Modeling - UESC (2025)
📜 Postgrad in DevOps Engineering - IFMT (in progress)
⚡ BEng in Electrical Engineering - UESC (2020)

Let's connect

🇧🇷 Versão em Português

Sobre mim

Engenheiro eletricista, mestre em Modelagem Matemática e Computacional e doutorado em andamento em Modelagem Matemática e Computacional (UESC). Trabalho na interseção entre machine learning informado por física, simulação numérica e operações orientadas a dados. Experiência liderando equipes, construindo pipelines de automação em Python e desenvolvendo modelos de previsão para planejamento de capacidade.

Busco aplicar minha base quantitativa forte e experiência com modelagem e ML em problemas reais de escala industrial.

Navigation Menu