The product
WeLoveJoe is a platform for building business-specific agents as configurable workflows. Agencies and teams can model how an agent asks, replies, calls tools, and moves through a process instead of commissioning a separate hard-coded bot for every use case.
I worked on the platform as a full-stack engineer.
Node-based composition
Workflows are assembled from nodes with defined responsibilities. Ask nodes collect information. Reply nodes control what the agent returns. MCP nodes connect the workflow to external tools and data. Additional nodes shape routing and the wider execution path.
This turns an agent from one large prompt into a system a business can inspect and change.
Business-specific behavior
The same platform can support different processes because behavior lives in the workflow definition. An agency can adapt the conversation, available tools, and response path for each client without rebuilding the runtime from zero.
Full-stack contribution
My work covered the authoring experience and the application behavior needed to execute those configured flows. The central challenge was keeping what users build in the interface aligned with what the agent runtime actually does.
The result
The platform gives teams a structured way to design agent behavior from smaller, explicit steps. Workflows are easier to reason about, reuse, and adapt than a collection of opaque prompts.
Stack
TypeScript · Node.js · Agent Workflows · MCP · LLM APIs · Full Stack