Wendor

Machine-status and order endpoints were slow on large datasets. The dashboard mixed operational and analytical work on the same pages, and unused queries rode along with the data that operators actually needed.

I restructured SQL on getAllMachineStatusV4 and getAllOrdersServiceV2 with window functions, parameterized filters, and parallelized lookups. Response time on those endpoints went from about 15 seconds to about 3 seconds.

On the dashboard I rewrote analytical queries, migrated the data layer to React Query, removed unused queries, and split operational pages from analytical ones. Load time improved 40–60%. I also wrote team guidelines so that pattern stuck.

Read traffic moved off the primary: client-facing read APIs now route to PostgreSQL replicas on AWS RDS, with a critical-read tier for high-priority endpoints. Separately I built PO-v2/GRN receiving, a warehouse product ledger, and snapshot inventory.

The 15s → 3s and 40–60% figures are owner-reported; Git does not independently measure them. A weekly_machine_stat materialized view existed from before my tenure and is not my work. Cloudflare was not part of Wendor. Mobile authorship on the field-ops app is not evidenced.