§01
Built WhatsApp chatbot systems on a proprietary MCP (Model Context Protocol–style) architecture, serving multiple tenants with context-aware responses while maintaining strict data isolation. That last clause is the hard part: a shared conversational AI layer has to keep every tenant's context, history, and data from leaking into another tenant's session. It's the same failure mode that turns a promising multi-tenant AI product into a liability.
§02
Refactored the order PDF payload build flow, eliminating 75% of unnecessary requests and improving service speed, which in turn reduced weekly bug reports for the service by 90%. The causal chain matters: cutting redundant requests wasn't just a performance win, it removed most of the race conditions and partial-state bugs the old flow was generating, which is why the bug-report number moved almost as much as the request count did.
§03
Engineered a multi-tenant ordering system integrated with ERPs, managing complex workflows, tax calculations, and financial data across more than 50 tenants. Tax logic and financial data are unforgiving of shortcuts: a bug here isn't a broken UI, it's a wrong invoice. So the system had to get workflow orchestration and ERP integration right simultaneously across every tenant's own configuration.
§04
Built a React-based administrative platform for onboarding and access control, which became the central interface for WhatsApp automation clients, managing more than 200 manager-level users. This is the operational control plane behind the tenant-facing chatbots: the place where access, onboarding, and configuration for every client organization actually gets managed.