LLM Agent Workflow Platform / AgentflowV2

Tech Lead · Platform owner · 2024

Owning the platform

As top technical contributor and de facto owner, I directed architecture across a Node.js/TypeScript monorepo deployed on Azure Container Apps — a low-code platform where teams compose LLM and agent workflows without writing orchestration code.

The centerpiece was AgentflowV2, the platform’s second-generation graph-based workflow engine: loop, iteration, and conditional agent nodes with live execution tracking, designed and delivered through to general availability.

What changed

Multi-tenant security hardening that closed cross-tenant data-leak vectors via org-scoped credential and data-label isolation.
Three LLM providers (OpenAI, Azure OpenAI, Gemini) integrated into a centralized embedding and document-store management layer.
Full three-language localization (English / Japanese / Vietnamese) and production-reliability hardening across UAT and prod.

Talk through a similar problem ↗