TradeSpark runs a B2B marketplace connecting hardware distributors with procurement teams across Southeast Asia. At peak, their platform handled 50,000 daily sessions across 12,000 SKUs, but the aging codebase was struggling to keep pace.
The migration strategy
We phased the migration over 14 weeks to avoid a big-bang cutover. The catalog, search, and checkout were extracted independently, each validated against feature parity before going live. Traffic was shifted using Cloudflare weighted routing, giving us instant rollback capability at every stage.
Performance architecture
Static generation handles 95% of catalog pages. The remaining 5% (real-time pricing, stock levels, and personalised bundles) are fetched client-side from a dedicated pricing API with aggressive edge caching. The result: a Lighthouse performance score jumping from 34 to 96.
Search that actually converts
We replaced keyword-only search with a semantic layer using pgvector. Buyers can now find products by description, application, or part number, and conversion from search to cart improved by 28% in the first month post-launch.
