Talks

Autoresearch · Link to this talk

An introduction to autoresearch as a way of turning a coding agent into a constrained experimentation system. Drawing on Andrej Karpathy's project, the talk explains how to limit what the agent can modify, evaluate each attempt with fixed metrics, and retain the best results reproducibly, as well as the risks of optimising a metric that does not represent the real objective.

Coding agents · Automated experimentation · Evaluation · Reproducibility · Python

Materials pending publication.

Optimizing Python and Rust · Link to this talk

A practical talk on software optimisation using the Longest Increasing Subsequence problem. It compares different algorithms and implementations in Python and Rust to show how algorithmic complexity, runtime, optimisation techniques, and compilation mode affect performance, using reproducible benchmarks and visualisations.

Python · Rust · Algorithms · Performance · Benchmarking