eCommerce Marketplace

A perfume retailer with 8K+ SKUs was bleeding customers through a storefront that couldn't keep up. The platform started as a single-vendor shop and got stretched into a marketplace through bolt-on integrations. Every seam was showing.

Product pages took 4-6 seconds to load on mobile. Every request hit the database directly with no caching, no CDN, no image optimization

Search was keyword-only. Typing "perfume" returned 200+ unranked results with no way to filter by brand, price, or availability

Cart abandonment sat at 62% because checkout required 5 separate page loads and re-entered shipping details every time

No guest checkout. Customers who didn't want to create an account just left

Wishlists didn't exist. Customers bookmarked browser tabs to save products for later

Mobile was a responsive afterthought: overlapping form fields, desktop-sized images on 4G, buttons too small to tap

The business wasn't losing to better products. It was losing to competitors with faster, smoother shopping experiences.

I rebuilt the marketplace from scratch as a mobile-first platform designed for high-volume browsing, fast checkout, and customer retention. Every decision optimized for the moment a customer decides to buy or bounce.

Built for thumbs first, mice second.

✅ Resolution-aware lazy-loaded images serving WebP at the correct viewport size, cutting image payload 60% on mobile

✅ Bottom-sheet category navigation instead of hamburger menus, reducing tap-to-browse from 3 taps to 1

✅ 95+ Lighthouse performance score with sub-2-second initial page loads on 4G connections

📊 Outcome: Mobile page loads dropped from 4-6 seconds to under 2 seconds. Bounce rate on product pages fell 38%

Customers find what they want in 2 keystrokes, not 20 scrolls.

✅ Full-text search with typo tolerance, synonym matching, and relevance ranking across 8K+ products

✅ Faceted filtering by category, price range, brand, rating, and availability updating in real time without page reloads

✅ Search suggestions after 2 keystrokes with product thumbnails and pricing inline

📊 Outcome: Search-to-product click-through rate increased 52%. Average time-to-first-product dropped from 45 seconds to 12 seconds

No more "sorry, out of stock" at checkout.

✅ Cart validates against live inventory on every interaction. Stock changes trigger immediate quantity adjustments with clear customer notifications

✅ Cross-sell suggestions based on cart contents and browsing history

✅ Dynamic shipping cost estimation updating as items are added or removed

📊 Outcome: Out-of-stock errors at checkout dropped to near zero. Cross-sell suggestions drove 8% increase in average order value

Turn window shoppers into repeat buyers.

✅ One-tap save with price change and stock status tracking on every wishlisted item

✅ Automated notifications when prices drop or out-of-stock items return

✅ Back-in-stock alerts drove 40% of returning user engagement within the first month

📊 Outcome: Price-drop notifications generated a 28% conversion rate on wishlisted items. Wishlist became the highest-engagement retention feature on the platform

5 pages → 1. Guest checkout. Saved payment methods. Done.

✅ Guest checkout with optional account creation post-purchase

✅ Saved addresses and payment methods for returning customers with one-click reorder from order history

✅ Returning customers complete checkout in under 60 seconds

📊 Outcome: Cart abandonment fell from 62% to 34%. Checkout completion hit 78% for returning customers

Why I chose this stack and what tradeoffs I made.

Next.js with ISR over full SSR — 8K product pages can't all be server-rendered on every request. Incremental static regeneration gives sub-1-second loads with SEO indexing while keeping build times manageable. Dynamic pages (cart, checkout) stay server-rendered

PostgreSQL full-text search over Elasticsearch — At 8K products, Postgres full-text with GIN indexes handles search in under 200ms. Avoided the operational overhead of a separate search cluster for a catalog this size. Faceted filtering runs against denormalized materialized views

Stripe tokenization for saved payments — Card-on-file with 3D Secure keeps PCI scope minimal. Webhook-driven order confirmation means payment failures never leave orders in limbo

CloudFront CDN for images — Product images served from edge locations with automatic WebP conversion. Origin S3 stores originals, CloudFront handles format negotiation per device. Cut image delivery latency 70%