RB

Projects

I reimplement recent ML papers from scratch to understand them end to end, then build real systems on top. These are the ones I'm most proud of, and many are open-source with live demos.

Snowy mountain landscape generated by the latent diffusion model

Latent diffusion from scratch

PyTorch

Rectified-flow latent diffusion, MMDiT-style, written from the paper up.

GITHUB ↗
Homegrown LLM text-generation playground

Homegrown LLM

PyTorch

A transformer LM pretrained from scratch, then instruction-tuned.

GITHUB ↗
NeuroZip, QBI Hackathon 3rd place winner3rd · QBI Hackathon

NeuroZip

PyTorch

3rd place at the QBI Hackathon: EEG compression you can search by text.

GITHUB ↗
Parameter Golf training log: 600s wallclock, int8+zlib 9.6MB artifact, sliding-window BPB ~1.51

OpenAI Parameter Golf

PyTorch

Best language model under 16 MB: 1.5126 bits/byte in 600 seconds.

GITHUB ↗
PyTorch

V-JEPA world model

PyTorch

A self-supervised video world model hand-written in PyTorch.

GITHUB ↗
OtterAgent brain-map interface with an agent chat panelBest AI Hack

OtterAgent

FastAPI · Gemini

A social network where AI agents negotiate on your behalf.

GITHUB ↗
Vesta recordings library with AI threat scores

Vesta, surveillance engine

Flask · YOLO · llama.cpp

Multi-camera theft detection with YOLO triggers and VLM threat scoring.

GITHUB ↗
btop showing M3 Max CPU and GPU usage on macOS

btop, with macOS GPU

C++

GPU telemetry in btop for Apple-Silicon PyTorch users, no sudo.

GITHUB ↗