My Work
The most important roles, projects, and ideas that I put my time into.
Toy vectorised autodiff engine implemented in Numpy, inspired by autograd; manual backprop and gradient descent computations supported. Optimised 2-layer MLP on MNIST to PyTorch equivalent performance.
Compositional AI systems final research project: bulding a decentralised memory architecture to promote decentralisation and scalability of multi-agent systems.
Software engineering w/ Generative AI final project: graph-based AI research framework for information extraction and divergence from Arxiv, OpenAlex, CORE, etc.
Biologically-focussed productivity tool prototype focussed on kairos productivity (see Tiny Experiments).
Graph-based knowledge management system focussed on quick capture and efficient retrieval of information. What started my obsession with personal knowledge management, graphs, context/the search problem, and AI agents (lwk everything I work on nowadays).