DiffusionLab is a desktop application that aims to democratize access to the latest open source diffusion models (namely Stable Diffusion and its variants) by providing a turnkey desktop application that allows end users to use their own hardware to generate and edit images. It already contains basic text-to-image and image-to-image support, with ongoing work to support inpainting, outpainting, and textual inversions. Its codebase also serves as a testbed for my own novel research on diffusion models.

It is currently closed-source as I may decide to commercialize it, but I am also considering releasing an open-source version.

Built using Electron, React, Node, TypeScript, and PyTorch.

I have ongoing research projects to extend Stable Diffusion’s capabilities.

Projects include:

• Novel high-fidelity inversion techniques that allow users to embed known objects/concepts into generated imagery.

• Novel guidance approaches.

• Domain-specific fine tunings.

• Extension of diffusion models to handle video generation.

All of my experiments are run using PyTorch. Most are run on Google Colab, but I also utilize Lambda Labs GPU cloud and GCP when I need something with more firepower.

Sandpiper is a business software built for the niche requirements of law schools participating in mock trial competitions. It manages data collection, scoring, calculation of competition results, reporting, and archival. It is being developed in collaboration with a long-time friend and colleague as part of Pentaclex.

As the lead and sole engineer on this project, I am fully responsible for the end-to-end implementation, testing, monitoring, and deployment.

Built using ASP.Net Core MVC and SQL Server, hosted on AWS.