Reyner Pantou
Architecture and migration / Professional work

A simpler search API layer

Migrated Search API to ByteCloud and separated Search Provider engine aggregation from Search Pack item and component packing.

Role
Technical Owner · API Layer
Delivery
Phased Migration & Full Rollout
Tech Stack
Go · Kitex · ByteCloud
Impact
−350 ms Search Result Page Latency

The Problem

The existing search service mixed API orchestration, engine integration, and product-card logic. Moving platforms also created an opportunity to separate responsibilities and make changes easier to test, deploy, and diagnose.

The Architecture

Search API orchestrates Search Provider engine aggregation and Search Pack item and component assembly. Organic and advertising engines are dependencies beyond the owned API layer.
A simplified view of service responsibilities. Highlighted services are the API layer; engine internals are outside this ownership scope.
01

Service boundaries

Separated Search API orchestration, Search Provider engine aggregation, and Search Pack item/component assembly.

02

Safe cutover

Planned migration steps and monitored search behavior across the affected surfaces.

03

Operational clarity

Used service-specific dashboards and debug tools to make ownership and issue diagnosis clearer.

The Execution

Owned re-architecture and cutover of the search API layer for the GCP-to-ByteCloud migration. The service boundaries were Search API for user-facing orchestration, Search Provider for engine aggregation, and Search Pack for item and component packing. I took the new services through stress testing and full rollout.

What I owned

  • Created the architecture, migration strategy, and cutover plan for the search API layer.
  • Served as main point of contact for Search Provider and Search Pack, implementing them and monitoring the rollout to 100%.
  • Coordinated QA and Ads engineering to check both organic and advertising behavior during the transition.

Technical impact

  • Reduced reported Search Result Page latency from 1,200 ms to 850 ms, a 350 ms reduction.
  • Built Search Provider for organic/ad engine aggregation and Search Pack for item and component packing, enabling pack features to deploy independently of Search API.
  • Created a shared service boilerplate, debugging utilities, dashboards, stress tests, and rollout monitoring. Performance records estimate two engineering days saved per engine or product-pack feature.

The Result

Search Result Page Latency
−350 ms

1,200 ms to 850 ms

API-Layer Infrastructure Cost
−54.68%

ByteCloud Migration & Service Improvements

Reported Cutover Downtime
0

Search service migration

Business impact

  • Reported annual infrastructure cost for the Search API layer fell by approximately $86,845. This is a combined before/after result of ByteCloud pricing, service improvements, and re-architecture; it does not isolate the savings from engineering changes.
  • Kept search product, shop-page product, and shop-query functionality available through the migration.

The Takeaway

The migration reduced operating cost while giving the team clearer service ownership and independent deployment paths.