Bangladesh's online commerce reads, today, as two extremes. There are single-brand sites — well-designed, narrow catalog, slow growth. And there are generic marketplaces that scale wide but lose the seller-verification promise the customer is actually paying for. Pentamart's bet was that there was a third lane: a verified-vendor marketplace where breadth and trust hold at the same time.

The product strategy followed from that. Customer-facing surfaces had to feel like one place. Backend systems had to model the vendor as a first-class entity with its own catalog, payouts, and approval workflow. And logistics had to bend so a customer ordering from three different vendors received one delivery — otherwise the marketplace just shifts cost onto the buyer.

The customer storefront is React on a Node.js / Express API, served from a single PostgreSQL schema that holds vendors, products, orders, and payouts. Neon hosts the database with branchable previews — every PR gets a clone of production for QA without touching live data.

Tailwind carries the design system. We built a small set of primitive components — product card, vendor strip, filter rail, combined-cart row — and composed every page from them. The grocery flow and the electronics flow share roughly 80% of their surface area; they just configure differently.

The combined-cart engine is the part the customer never sees. A single order is decomposed into per-vendor fulfilment jobs at checkout, each tracked against the vendor's stock and shipping window. A logistics layer then re-merges the parcels at the destination hub before final delivery. The customer pays once, signs once, receives once.

The first version of vendor onboarding was a six-step wizard. Sellers stalled. We collapsed it to two screens — identity verification, then a catalog upload — and moved the rest of the metadata to ongoing edits in the vendor console. Time-to-first-listing dropped sharply.

Weekly Thursday flash sales (7–9 PM) became a fixture early on. The platform now treats them as a first-class campaign type — vendor opt-in, automatic price banners, time-bound stock allocation — rather than a manual one-off. Revenue concentration around those two hours is meaningful and predictable.

Five years in, the next rooms to unlock are post-purchase: vendor ratings, returns automation, and a subscription mode for groceries. The architecture — vendors as first-class entities, orders as event streams, logistics as a separate concern — generalises cleanly to all three.