4 projects
veil-pii
Reversible PII masking for LLM calls: mask before the prompt leaves your network, restore in the answer, streaming included.
flakemap
Find the flaky tests in your CI history from JUnit XML reports.
trainspotter
Automatic diagnosis of loss curves and training logs: spot divergence, spikes, overfitting, LR and throughput problems before you burn another GPU-day.
errorbars
Error bars for LLM evals: standard errors, clustering, paired comparisons, leaderboards, and power analysis.
