AI Workbench
An internal AI assistant built on Gemini that lets company teams query business knowledge
and pricing rules directly, cutting down on manual lookups and reliance on tribal knowledge.
My role: backend integration — built the custom retrieval framework that feeds company
SOPs and pricing rules into Gemini and exposed the assistant to internal teams.
Replaced near-daily one-off IT requests from the Category and Channel teams with a
self-service AI assistant built on a self-made context-retrieval pipeline (no vector
database), giving non-technical staff direct access to pricing rules and company
knowledge without needing database access or engineering support.
Node.js | TypeScript | Express | Gemini API | Custom Retrieval Framework
Pricing Management Platform
A pricing engine that keeps prices, inventory-aware recommendations, and promotions in sync
across 35+ sales channels, including Amazon AU, Woolworths MarketPlus, and eBay.
My role: designed and built the pricing engine and promotion framework across a 14-service
monorepo, using AI-assisted development (Claude Code with Claude Sonnet) to accelerate
delivery, and owned performance tuning end to end.
Optimized the high-frequency pricing engine, slashing core sync workflow latency by 80%
through query, indexing, and caching improvements, while resolving critical concurrency
bottlenecks so channels never serve stale prices.
Node.js | TypeScript | Express | Prisma | PostgreSQL | Monorepo (14 services) | Claude Code (Claude Sonnet)
Channel Management System
A master-data platform for channel profiles, commission structures, and sales analytics.
My role: independently designed and delivered the system end to end, using AI
pair-programming (Cursor with GLM) to speed up delivery, including the CI/CD pipeline.
Consolidated previously scattered channel contacts, ownership details, and integration
info with category-level sales data into a single system, giving leadership clear
visibility into channel performance and ownership.
TypeScript | Express | Prisma | PostgreSQL | React | Vite | Docker | Cursor (GLM)
Zendesk Integration
Middleware connecting order management and 40+ e-commerce/dropship channels to Zendesk for
unified support handling.
My role: designed the integration architecture and delivered 50+ REST API endpoints for
ticket creation, updates, merging, search, and export, using GitHub Copilot's autocomplete
to speed up boilerplate and endpoint scaffolding.
Built a tag-based classification layer to unify chat, chatbot, and email messages from 40+
channels into one ticket pipeline, pre-fetched channel/customer data so agents weren't
waiting on live lookups, and worked directly with support and ops teams to shape routing
rules around how they actually triaged tickets.
Node.js | Express | Zendesk | REST API | Integration Middleware | GitHub Copilot
After-Sales Management System
A full-lifecycle after-sales platform for a multi-department e-commerce operation, spanning
ticket creation, multi-level approval, refund execution, and warehouse return processing.
My role: owned the backend architecture end to end and translated requirements from
finance, warehouse, and support teams directly into the approval workflow.
Implemented PayPal/bank-transfer refund pipelines with financial checkpoints and MongoDB
full-text search plus exportable reporting, enabling the support team to handle 25% more
tickets daily with the same headcount.
PHP | Zend Framework | MySQL | MongoDB | Google Drive API | REST/OAuth
Warehouse Return APP
A return-processing web app covering login, barcode/tracking scanning, photo evidence, and
submission end to end.
My role: built the frontend and the OMS/SMG integration layer that looks up and merges
order/return data.
Added camera-based SKU recognition and validation logic across four return paths
(restock, second-hand, dispose, QC/NCR), giving warehouse staff instant product
identification and routing decisions that cut manual processing time and increased
daily throughput.
React | Redux Toolkit | Axios | IBM Carbon Design System | OMS/SMG APIs
