Things I have built
Code that actually runs. Most of these started as coursework or side projects and turned into something worth sharing.
Generative modeling
Diffusion · PyTorch · Sampling · FID
Variance-Preserving Diffusion for Image Generation
A VP diffusion model written from scratch: custom noise schedules, forward/reverse processes, and few-step sampling with NFE 1, 2, and 4.
Compared consistency-model and ReFlow-style shortcuts to see what actually lowers FID when you only get a handful of function evaluations.
Open repositoryComputational neuroscience
Wilson-Cowan · Dynamical systems · Stochastic processes
Stress, Delay & Noise in Gamma-band Oscillations
A Wilson-Cowan circuit that shows how tonic drive, feedback delay, and noise reshape cortical rhythms.
The surprising result: moderate noise increases coherence. The model also reproduces gamma-to-beta shifts seen in stressed cortex.
Open repositoryApplied ML
Computer vision · YOLOv8 · CLI · OpenCV
Real-Time Object Counting (YOLOv8 + OpenCV)
A CLI that counts objects in images, video files, YouTube links, or a live webcam feed.
Modular enough to swap detectors. Tuned to run on CPU (macOS) without feeling slow.
Open repositoryMore explorations
Generative modeling
Diffusion Image Generation Challenge
Pitted diffusion against flow-based methods to see which generates better images in under four sampling steps.
Most of the gains came from scheduler design and denoising tricks, not architecture changes.
Diffusion · Flows · Image synthesis
Open repositoryML foundations
ML & Generative Modeling Notebooks
End-to-end experiments: data splits, bias-variance, logistic regression, feedforward nets, neural ODEs, and denoising score matching.
Every notebook is self-contained and designed to be re-run without setup pain.
ML pipelines · Neural ODEs · Statistics
Open repositoryBiological modeling
Mathematical Biology Mini-Projects
ODE models, parameter estimation, clinical stats, and single-cell analysis across several small projects.
Mixed dynamical systems with dimensionality reduction (PCA, t-SNE, UMAP) to make sense of real data.
ODEs · Optimization · Single-cell data
Open repositoryTheoretical neuroscience
Neural Dynamics & Information Processing
Neuron models, spike train analysis, STDP, and information-theoretic views of how neurons encode signals.
The point was to explain behavior with simple models, not just fit outputs.
Neural coding · STDP · Simulation
Open repository