Applied Mathematician & Machine Learning Engineer
Bridging pure mathematics, computational science, and deep learning.
🌐 Personal Website & Blog • 🚀 Projects Portfolio • 📄 Resume (EN) • 📄 CV (ES)
- 🎓 Senior Mathematics Student at Universidad Yachay Tech (Ecuador).
- 🤖 Machine Learning Engineer trained via intensive bootcamps at Anyone AI.
- ☁️ Certified Cloud Application Developer candidate at AWS Cloud Institute.
- 🔬 Research focus: Scientific Machine Learning (SciML), Physics-Informed Neural Networks (PINNs), and Computer Vision for biomedical applications.
- ✍️ Author of scientific essays on history, mathematics, and great minds at jeffersonconza.github.io/blog.
- Physics-Informed Neural Networks (PINNs): Solving nonlinear PDEs (Burgers 1D, 2D Heat Diffusion, Navier-Stokes) by embedding physical conservation laws directly into neural network loss functions.
- Biomedical Computer Vision: Isotropic 3D image reconstruction via generative models (CycleGAN) to resolve axial resolution anisotropy in fluorescence microscopy.
- Microplastics Detection: Environmental ML classification and regression modeling using FTIR spectroscopy.
- Cloud & MLOps Architecture: Deploying containerized inference engines with FastAPI, Docker, and AWS infrastructure (EC2, S3).
|
Modular PyTorch and DeepXDE framework to solve Burgers, Heat, and Navier-Stokes equations with physics losses. Automated checkpointing to AWS S3 and containerized API on Docker/EC2.
|
Hybrid Deep Learning architecture combining Swin Transformer and DenseNet121 for pediatric pneumonia screening in chest X-rays. Achieved 99.0% recall with Grad-CAM explainability.
|
|
Unpaired image-to-image translation with 3D CycleGAN to restore spatial resolution along the axial (z) dimension in confocal microscopy, eliminating PSF blur.
|
End-to-end time-series energy demand forecasting platform using XGBoost, Random Forest, and Ridge Regression. Microservice architecture with FastAPI and interactive Streamlit UI.
|
|
Machine Learning classification and spectral feature extraction for environmental microplastic pollutants using Fourier-Transform Infrared (FTIR) spectroscopy.
|
Programmatic mathematical and physical animations crafted with Manim for educational outreach, orbital mechanics, and scientific storytelling (Gagarin Day).
|
- Core & Scientific Computing: Python, Julia, R, SQL, C++, Bash, LaTeX
- Deep Learning & SciML: PyTorch, DeepXDE, TensorFlow, scikit-learn, XGBoost, Hugging Face, Torchvision
- Math & Data Engineering: NumPy, pandas, SciPy, SymPy, Matplotlib, Seaborn, Manim
- MLOps & Cloud: Docker, FastAPI, Streamlit, AWS (EC2, S3, SageMaker), Git, Linux / Unix
import torch
# Embedding physical differential operators into the loss function
def physics_loss(u_net, x, t, nu=0.01 / torch.pi):
x.requires_grad_(True)
t.requires_grad_(True)
u = u_net(torch.cat([x, t], dim=1))
# Automatic differentiation (autograd)
u_t = torch.autograd.grad(u, t, grad_outputs=torch.ones_like(u), create_graph=True)[0]
u_x = torch.autograd.grad(u, x, grad_outputs=torch.ones_like(u), create_graph=True)[0]
u_xx = torch.autograd.grad(u_x, x, grad_outputs=torch.ones_like(u_x), create_graph=True)[0]
# 1D Viscous Burgers' equation residual: f = u_t + u * u_x - nu * u_xx
residual = u_t + u * u_x - nu * u_xx
return torch.mean(residual ** 2)"PyTorch turns mathematical intuition into tensors, autograd into differential equations, and neural networks into computational physics engines."
