Saiful Khan

Building the future through software

Initializing portfolio0%
Skip to content
All projects

Commerce Infrastructure · 2025

Atlas Commerce

Headless commerce platform serving multi-region catalogues at edge latency.

Product visual
Revenue lift
+31%
Infra cost
-42%
Deploy frequency
12x

01 — Overview

What this system is

A headless commerce core replacing a monolithic storefront for a multi-brand retailer operating across nine markets.

02 — Challenge

The constraint that mattered

Catalogue reads were blocking checkout under peak load and every market required its own pricing, tax and content rules.

03 — Research

How the problem was measured

Traffic replay across three peak events isolated the read path as the dominant tail-latency contributor.

04 — Solution

The approach taken

A read-optimised edge catalogue with regional pricing resolution, decoupled from a transactional core that owns checkout.

05 — Architecture

System composition

  1. 01Edge catalogue cache with per-market invalidation
  2. 02Transactional core on Postgres with row-level tenancy
  3. 03Event bus for inventory and price propagation
  4. 04Composable storefront rendered at the edge

06 — Development

Delivery model

Delivered in fortnightly increments behind flags, with each market cut over independently and rollback available per region.

  1. Discovery

    Traffic replay, latency attribution, market rule audit.

  2. Architecture

    Read/write split, tenancy model, invalidation contract.

  3. Build

    Edge catalogue, pricing engine, storefront blocks.

  4. Cutover

    Per-market migration with instant rollback.

export async function resolvePrice(sku: string, market: Market) {
  const cached = await edge.get(`price:${market}:${sku}`)
  if (cached) return cached

  const price = await core.pricing.resolve({ sku, market })
  await edge.put(`price:${market}:${sku}`, price, { ttl: 60 })
  return price
}

07 — Features

What shipped

  • Multi-market pricing engine
  • Composable storefront blocks
  • Realtime inventory sync
  • Checkout with idempotent order writes
  • Merchandising preview environments

08 — Stack

Technology

  • Next.js
  • TypeScript
  • PostgreSQL
  • Redis
  • Cloudflare
  • Docker

09 — Performance

Measured in production

LCP
0.9s
TTFB
82ms
Peak RPS
14k
Error rate
0.02%

10 — Search

SEO outcomes

Indexed pages
412k
Organic sessions
+164%
CWV pass rate
100%

11 — Gallery

Interface surfaces

  • Storefront
  • Merchandising console
  • Pricing engine
  • Observability

12 — Impact

Business outcome

Checkout conversion recovered during peak trading and merchandising cycles moved from weekly to same-day.

Building something comparable?

Start a conversation