Agentic AI / Product engineering
Shadow PM.
A multi-agent platform that brings Jira, customer transcripts, and codebase context together for product research and UI prototyping.
The problem
Product decisions draw on scattered context: customer conversations, existing tickets, and what the codebase actually supports. Turning that information into a useful plan or prototype takes both research and coordination.
What I built
Shadow PM connects those sources in a multi-agent platform for product research and UI prototyping. I built it with a Python / FastAPI backend, Temporal orchestration, and LangSmith tracing.
The technology in context
- Python and FastAPI provide the backend for the platform and its integrations.
- Temporal orchestrates the multi-step agent workflows.
- LangSmith provides tracing to inspect how the AI workflow runs.
- Jira, customer transcripts, and codebase analysis supply context for research and prototype generation.
The homepage diagram illustrates the flow from context through specialist agents to a prototype.
The engineering focus
The work goes beyond calling an LLM: connecting useful context, coordinating several steps, and making execution inspectable. The result is a workflow that turns product research into something a person can review and refine.