Projects

Selected research projects in mechanistic interpretability, machine learning, deep learning, and high-performance computing.

Mechanistic Interpretability
project thumbnail

Implicit Personalization

Detecting and attributing the user models that LLMs build implicitly during conversation
project thumbnail

MoE Interpretability

Reading MoE experts with sparse vocabulary directions and testing when specialization is real

Machine Learning
project thumbnail

BayesianFlow

Pixel-wise uncertainty estimation in Flow Matching generative models via Last Layer Laplace Approximation
project thumbnail

LoRA & DoRA in TinyGrad

From-scratch Low-Rank Adaptation and Weight-Decomposed LoRA implemented in TinyGrad
project thumbnail

Genetic Algorithms for Neural-Network Pruning

Evolving structured pruning masks for a FashionMNIST MLP while keeping its trained weights frozen

HPC
project thumbnail

Self-Attention Kernels

Optimized causal multi-head self-attention in CUDA, OpenMP, and SIMD, reaching 1.09× naive PyTorch on A100
project thumbnail

Parallel Heat Stencil

A hybrid MPI and OpenMP heat stencil scaled from one CPU core to multiple Leonardo nodes