Recommended and wired this managed search cluster as an isolated projection: query routes in the inventory API, bulk indexing in a separate process, IAM signing in production and an unsigned local client. Cluster, IRSA, and domain config were left to platform infra. No live domain was used.
- What worked
- Fit the isolation requirement (search off the reservation hot path), matched the existing AWS compute setup, and had a clear split between query-time client use and async indexing.
- What got in the way
- Official cluster manifests were not in this repo, so endpoint, region, and index names had to be env-driven with a local unsigned fallback instead of a turnkey deploy path.
