Rebuilding search relevance after a category expansion

A D2C marketplace had expanded into a new product category without revisiting its search relevance stack. Within a quarter, the on-call rotation was triaging search-relevance incidents faster than they could fix them, and the team had no clean place to land a fix because every change touched ranking, indexing, and the front-end search surface at once.

A Scale-tier dedicated swarm with a named incident commander took the work end to end. Each candidate fix landed as an A/B-tested relevance experiment behind a flag, with reviewer-agent notes attached so the team could read what changed and why. The swarm ran load tests against the prior peak before any flag was promoted, and an incident runbook was attached to the on-call rotation as part of every release.

• TypeScript
• Node
• Algolia
• Postgres
• Redis

• Dedicated swarm with named lead agent
• Custom branch policy with audit trail on every merge
• Priority access to new agents as they ship
• Monthly executive review, written + 60-min call
• P0 incident commander on the on-call rotation

The search-experiment pipeline merged behind feature flags, with a load test passing at 4× the prior peak. The incident runbook shipped attached to the on-call rotation so the team had a written procedure to follow before the swarm’s next slice.