The architecture of the vector-quantized variational autoencoder lends itself well to non-linear clustering of complex datasets. I wrote code in PyTorch to find archetypal protein neighborhood structures, as seen in my co-authored paper in Bioinformatics.
Identifying multiscale structure in complex systems is crucial in building an understanding of agential, controlled behavior across organic and synthetic forms of life. Working toward a senior thesis, I implemented the mathematical frameworks proposed in Erik Hoel's Causal Emergence 2.0.
I was a primary author on this paper in Bioinformatics. Read more here, and find the codebase here.
From my longtime passion for Star Trek...
A memory-logging interface with cosine-similarity search, local transcription, AI integration, and more. Captain's Log helps you consolidate memories and recall how you processed events as they happened.
I self-host my data and have skills with the supporting tools: TrueNAS SCALE, Docker, etc. This box also acts as the host for my Arduino-controlled irrigation system (below).
Soil-dampness monitors automatically control the flow of water to my plants, with cameras that track their health and growth.