Automated CI and Validation Pipeline

Algorithm changes were being validated by hand, which made regressions easy to miss and made it hard to say whether a new revision was actually better than the last one. I built a pipeline that runs validation automatically on every change: Python for the validation logic, GitHub Actions for orchestration, and a Vercel and Supabase dashboard so the team could see results without reading logs.

The dashboard, /algo, is a QA-facing view of evaluation KPIs across algorithm versions and datasets, adopted by the quality team. Behind it: a 7-table schema (exe_versions, datasets, runs, run_datasets, run_results, trial_results, translations), RLS policies, an ingest API secured with a secret header, and migration safety reviews.

The design goal was making correctness visible to people who were not going to open a terminal. Engineers get run-level detail; everyone else gets a status view.