Stuff I've Made
Research reproductions, AI tools, experiments, and notes.
Majorana zero modes
Self-consistent Hartree-Fock for a quantum-dot–superconducting-nanowire hybrid.
Phys. Rev. BarXiv
Language-driven Segmentation
From-scratch DPT encoder + CLIP text encoder in a shared multimodal latent space (LSeg, ICML 2022).
Paper reproduction
MCTS for Connect 4
A playable game with an AI opponent powered by Monte Carlo Tree Search and UCB.
Prototype
Soil Organic Carbon with GNNs
Worked with Spatialise to predict soil organic carbon across 200K+ hectares with graph neural networks.
Recommendation Systems
Recommender models predicting the right service provider per user, on Yape’s 18M-user platform.
Distill
AI that turns any URL into a full progressive course with lessons, exercises, and solutions.
Prototype See output ↗
Neural Style Transfer
Artistic texture extraction via Gram matrices and VGG-19 features (Gatys et al.).
Paper reproduction
IonogramNET
Encoder-decoder CNN for extracting echo traces from Jicamarca Observatory ionograms.
Paper reproduction
Holographic Entanglement Entropy
Holographic entanglement entropy and Ryu-Takayanagi in AdS/CFT.
Thesis
Quantum Communication Complexity
Undergraduate research at USP on reduced-communication quantum protocols and the efficiency gains from entanglement.
Research
PPCA: The Minimal Generative Model
Deriving probabilistic PCA from a linear-Gaussian latent model.
Writing
Representing Atomic Environments for ML Potentials
How local atomic environments are represented in ML potentials.
Writing
Measuring the non-Markovianity of a Quantum System
Measuring non-Markovianity via decoherence rates and trace distance.
Writing
