Open Source

Recurring friction in day-to-day research work — dataset versioning, experiment comparison, quick embedding visualization — rarely justifies a heavyweight tool, but is still worth solving properly.

Build the smallest useful tool for each recurring problem, document it well, and release it rather than let it stay a private script.

A collection of independent, single-purpose packages rather than one large framework — each one easy to read end to end.

• Several tools adopted by former colleagues for their own ML workflows.
• Maintained with issue triage and periodic releases.

Consolidating documentation across the collection and adding a small set of shared utilities where genuinely useful.