Physics · Math · ML

Camilo Rozo

Machine Learning Engineer

Physicist and mathematician applying statistical mechanics, neural networks, and complex systems to real-world problems. 2+ years deploying ML models in production.

scroll
01 · About

About me

Physicist and Mathematician from Universidad Nacional de Colombia, focused on complex systems, statistical mechanics, and applied machine learning. I work at the intersection of physics, advanced mathematics, and software engineering.

Driven by the intellectual challenge of extracting information from seemingly indecipherable systems — high degrees of freedom, nonlinear dynamics, noisy observations. Applying mathematical and physical models to solve real-world problems and optimize human well-being and sustainability.

My undergraduate Physics thesis investigated diffusion processes on dynamic graphs, connecting network topology with phase transitions through statistical mechanics. That same approach — seeing the mathematical structure behind the data — goes into every ML project.

Every model is validated against exact analytical results, public benchmarks, or objective metrics. A model without rigorous validation isn't science — it's decoration.

Spanish
Native
English
Full Professional
2+ yrs
ML Engineer in production
2
Degrees at UNAL (Physics + Mathematics)
5
Portfolio projects
4
Certifications (ICTP · Stanford · UNAL)
02 · Experience

Professional experience

Trajectory in applied ML and academic tutoring at UNAL.

Feb 2024 — Present · 2+ years

Machine Learning Engineer

BS Energy And Automation SAS. · Part-time · Bogotá, Colombia · On-site
  • Develop and deploy production-ready ML models in Python, applying stochastic modeling and statistical mechanics principles to optimize large-scale network performance and predict systemic shifts.
  • Design and validate advanced algorithms for high-dimensional datasets, translating theoretical frameworks into practical solutions for infrastructure modernization and distributed system efficiency.
  • Implement rigorous model-tuning strategies with deep error analysis and hyperparameter optimization to ensure high-fidelity predictions in nonlinear environments.
  • Streamline model performance for deployment, balancing computational complexity with hardware constraints to enable efficient real-time processing without loss of accuracy.
Aug 2022 — Dec 2025 · 3 yrs 5 mos

Mathematics Tutor

Universidad Nacional de Colombia · Part-time · Bogotá · Hybrid
  • Academic support to students in physics and mathematics courses across 6 consecutive semesters.
  • Developed communication skills for simplifying highly abstract theoretical and mathematical concepts to diverse audiences.
Best Mathematics Tutor · 2024
03 · Education

Education

Double degree in Physics and Mathematics at Universidad Nacional de Colombia.

Feb 2018 — Dec 2024

B.Sc. in Physics

Universidad Nacional de Colombia

Undergraduate thesis on complex systems and diffusion processes on dynamic graphs, using statistical mechanics tools to connect network topology with phase transitions.

Feb 2018 — Dec 2025

B.Sc. in Mathematics

Universidad Nacional de Colombia

Proficient in Python and MATLAB for numerical computation and simulation. Integrating AI tools into the workflow. Favorite areas: topology and measure theory.

Feb 2010 — Dec 2017

High School Diploma · Science

Liceo Campo David

Recognized for the highest ICFES score in the graduating class.

04 · Certifications

Certifications

Complementary training in ML, complex systems, and advanced mathematical physics.

ML
April 2026

Supervised Machine Learning: Regression and Classification

DeepLearning.AI · Coursera · Stanford CPD · UVM
Credential ID: 7IWJT9H7I15S
Linear Regression Logistic Regression Gradient Descent Regularization scikit-learn NumPy
ICTP
March 2026

Spring College in the Physics of Complex Systems

Abdus Salam International Centre for Theoretical Physics (ICTP)
Statistical Mechanics ML Algorithms Complex Systems
UNAL
August 2024

ENREDANDO 2024 — Iberoamerican School of Networks and Complex Systems

Universidad Nacional de Colombia
Complex Systems Neural Networks Networks
UNAL
January 2024

Geometric Aspects in Physics and Mathematics

Universidad Nacional de Colombia
Differential Geometry Cosmology Math Physics
05 · Skills

Stack & competencies

Actively mastered and present across portfolio projects and production work.

λ

Machine Learning

Deep learning, Bayesian inference, autoencoders, Neural ODEs, PINNs, stochastic modeling.

PyTorchPyMCscikit-learntorchdiffeq

Applied Mathematics

Algebraic topology, measure theory, probability, analysis, optimization, ODEs/PDEs.

MCMCHomologyAutogradBayes

Physics

Statistical mechanics, phase transitions, dynamical systems, Monte Carlo, complex systems.

IsingMetropolisGraphsPDEs
</>

Engineering

APIs, containers, testing, Git, reproducible deployment, interactive demos, production tuning.

PythonMATLABFastAPIDockerStreamlit
06 · Projects

Projects

Five projects ordered by increasing complexity. Click any card to see the full detail.

Completed
01
Statistical mechanics · Bayesian

Bayesian Inference on the 2D Ising Model

Metropolis simulator with Numba for a lattice of spins, plus a PyMC model that infers the critical temperature and critical exponent via NUTS. Validated against the exact Onsager (1944) solution.

PythonPyMCNumPyArviZMonte Carlo
View details →
Completed
02
Deep learning · Dynamical systems

Neural ODEs for dynamical systems

An MLP parametrises the right-hand side of an ODE and is integrated end-to-end with a differentiable Dormand-Prince solver. Learns the vector field of a damped pendulum and a Lotka-Volterra system from short noisy trajectories, then is rolled forward beyond the training horizon to test extrapolation.

PyTorchtorchdiffeqODEsAutograd
View details →
Completed
03
Deep learning · MLOps

Anomaly Detection API · Autoencoder + FastAPI + Docker

LSTM autoencoder trained on healthy CMAPSS turbofan windows; reconstruction error becomes the anomaly score, calibrated against a held-out healthy validation set. ROC-AUC 0.969, recall 93.7 % at the q99 threshold. Served as a typed FastAPI service with a Streamlit demo, two-service Docker stack.

● Live demoPyTorchFastAPIDockerStreamlitPandas
View details →
Planned
04
ML for science · PDEs

Physics-Informed Neural Networks (PINNs)

Implementation of PINNs (Raissi et al. 2019) to solve the heat equation and the Burgers equation, validated against analytical and numerical reference solutions.

PyTorchPDEsAutogradResearch
View details →
Planned
05
Applied topology · Finance

TDA on financial time series — regime change detection

Persistent homology over sliding windows of multi-asset returns to detect transitions between normal regime and market crises (2008, 2011, 2015, 2020). Topological features as input to a supervised classifier.

Giotto-TDAPersistence Homologyscikit-learnPandasyfinance
View details →
07 · Contact

Get in touch

Let's Work Together

Open to Machine Learning Engineering, research engineering, and technical consulting opportunities. Particularly interested in projects at the intersection of ML, physics, and mathematics — complex systems, Bayesian inference, and applied topology.

Phone
(+57) 300 551 6912
Location
Bogotá, Colombia