Applied Mathematics & Computer Science undergraduate at Universidad del Rosario, Bogotá.
I build simulations, learning experiments and scientific software, then test their claims with strong baselines, held-out evaluation and explicit controls. My experience spans reinforcement learning, recurrent networks, spectral representations and retrieval-augmented LLM systems. My research interests are reasoning agents, optimisation, formal verification and computational neuroscience.
I bring applied mathematics and research engineering together: building an experiment is the starting point; understanding what its comparison establishes is the research question.
CV (PDF) · University email · Project guide
| Project | My work and what to inspect |
|---|---|
| SCRES: simulation and reinforcement learning | Research with Alexander Garrido: supply-chain simulation, Gymnasium/PPO experiments and matched-comparator evaluation. Start with the same-contract comparison. |
| Motor-RNN connectivity | Neuromatch Computational Neuroscience team project, followed by my independent equal-plasticity control. The control note separates structural density from trainable connections. |
| Spectral analysis of cognitive labor | My reanalysis of Andrade-Lotero and Goldstone's human-search experiment. The temporal audit reconstructs past-only features and compares spectral and coordinate representations; predictive advantage is not established. |
- HeliOS: a collaborative educational RISC-V kernel in C to which I contributed; architecture notes and QEMU smoke tests.
- ContratIA Abierta: open-procurement data pipelines, APIs and traceable signals for human review, not automated allegations.
- ChaosLab: a course project on double-pendulum dynamics, numerical checks and interactive explanation.
- Witty — LLM retrieval systems: built a professional FastAPI prototype with embeddings, optional FAISS search and LLM rewriting, and heuristic answer/escalate routing. This is retrieval and software-engineering experience, not model fine-tuning or calibrated uncertainty guarantees.
- Numerical analysis: implemented root-finding and quadrature methods in Python with analytical cases and convergence/error checks as coursework.
- Marmot social networks: observation-aware dyadic tables and a NetLogo/R model audit with Python and native-runtime checks, with Adriana Maldonado-Chaparro.
These entries are described in the CV, not offered as public reproductions. Private code, collaborator data and professional materials stay private.
Tools I use: Python, PyTorch, NumPy/SciPy, NetworkX, SimPy, Gymnasium, R, C, SQL, FastAPI, Git and pytest.
The linked PDF is the public distribution copy of my CV. Its editable source and version history are maintained separately; this profile is the guide to the work, not a second CV.