Webline Store
ProjectsWebline Store
AI-powered e-commerce at the edge
Overview
A production-grade e-commerce storefront built with Next.js 16 and deployed on Cloudflare's edge infrastructure. Features AI-powered product recommendations, multi-currency pricing (USD/KES), and a resilient microservices architecture.
Architecture
| Layer | Technology |
|---|---|
| Frontend | Next.js 16, React 19, Tailwind CSS 4 |
| Backend | Hono + oRPC (type-safe RPC) on Cloudflare Workers |
| Database | Cloudflare D1 (SQLite) + Drizzle ORM |
| AI/ML | Workers AI (BGE embeddings) + Vectorize |
| Caching | Multi-layer: Edge KV, Next.js Cache, TanStack Query |
| Payments | Stripe + M-Pesa (Daraja API) |
Hybrid Recommendation Engine
The recommendation system combines three AI strategies:
Content-Based Filtering
- 768-dimensional vector embeddings using Cloudflare Workers AI (BGE model)
- Semantic similarity computation for product relationships
- Pre-computed similarity matrices updated nightly
Collaborative Filtering
- User behavior tracking (views, cart additions, purchases)
- "Frequently bought together" and "Customers also viewed" patterns
- Event-driven updates via Cloudflare Queues
Context-Aware Scoring
Weights dynamically adjust based on user location:
| Context | Content | Collaborative | User Pref |
|---|---|---|---|
| Product Page | 60% | 30% | 10% |
| Homepage | 20% | 30% | 50% |
| Cart | 0% | 70% | 30% |
Customer:
Webline Technologies Ltd.Period:
6 monthsPerformance:
Full-Stack DevelopmentAI/ML IntegrationE-commerce PlatformPayment IntegrationEdge Computing
Features:
AI-Powered RecommendationsMulti-Currency Pricing (USD/KES)Real-time Cart SynchronizationStripe & M-Pesa PaymentsEdge-First Architecture25+ Database Tables
Primary Blue
Success Green
Warning Amber
Dark Gray
Challenge: Personalized Recommendations at Scale
The Problem
Building a recommendation system that delivers personalized product suggestions in real-time while handling thousands of concurrent users—without the cost and complexity of traditional ML infrastructure.
The Solution
Implemented a hybrid AI pipeline running entirely on Cloudflare's edge:
- Embedding Generation: Nightly batch job using Workers AI (BGE model) to generate 768-dimensional vectors for all products
- Pre-computed Similarities: Top 50 similar products per item stored in D1, eliminating real-time computation
- Context-Aware Scoring: Algorithm adjusts weights based on page context (product page favors content similarity, cart favors collaborative filtering)
- Sub-50ms Response: All recommendation queries served from edge cache with background revalidation
Challenge: Multi-Currency E-commerce
The Problem
Supporting both international (USD) and local Kenyan (KES) customers with accurate pricing, while handling currency-specific payment methods.
The Solution
Designed a dual-pricing schema with intelligent fallback:
- Base Currency: All products priced in USD as the canonical price
- KES Override: Optional local price that takes precedence when set
- Payment Routing: Automatic gateway selection (Stripe for USD, M-Pesa for KES) based on user preference
- Transaction Integrity: Atomic order creation with price locking to prevent race conditions
Challenge: Real-Time Cart Synchronization
The Problem
Users expect seamless cart experiences across devices and sessions, including merging anonymous carts when they sign in.
The Solution
Implemented event-driven cart architecture:
- Dual Identity Support: Carts track both
userId(authenticated) andanonymousId(cookie-based) - Smart Merging: On authentication, anonymous cart items merge into user cart with conflict resolution
- Real-Time Updates: Cloudflare Queues process cart events asynchronously
- Abandoned Cart Detection: Scheduled job (every 30 minutes) identifies stale carts for recovery emails
Gerald transformed our e-commerce vision into a reality that exceeded all expectations. The AI-powered recommendation engine he built has noticeably increased our customer engagement, and the seamless integration of both Stripe and M-Pesa payments opened our platform to both international and local Kenyan customers. His attention to performance—achieving sub-50ms response times on our edge infrastructure—shows a level of technical excellence that's rare to find.
Patrick Gitau
Chief Executive Officer · Webline Technologies Ltd.
