BBriuwu

Work project

Prosperna Commerce Copilot

At Prosperna, I work on Max, the merchant-facing AI copilot embedded inside a live commerce SaaS product. The system pairs a general orchestrator with specialist agents, streams responses into structured UI states, and applies hard product guardrails so merchants can use AI for real storefront work without losing control.

Back to projects

Overview

Project frame

Role

Applied AI engineering, streaming UX, workflow design

Timeline

June 2025 - Present

Status

Merchant-facing production AI system

Stack

  • React
  • TypeScript
  • Python
  • FastAPI
  • Server-Sent Events
  • Redis
  • Multi-agent orchestration
  • Approval workflows

Case study

What defines it

Product frame

Max is not a single feature. It is an orchestrated AI surface for merchant workflows inside Prosperna.

  • The general copilot handles branding and media tasks, then routes deeper work into specialist agents.
  • Specialists include Max Insights, Max Merchandiser, Max Copywriter, and Max Builder.
  • The experience is embedded in production UI, so the work has to be reviewable, stable, and understandable for merchants.

Interaction model

The core experience is a streaming chat that can open structured UI surfaces instead of only returning text.

  • SSE events drive widgets, approvals, memory moments, pickers, and artifact updates in the frontend.
  • Conversation restore and agent switching stay aligned through explicit active_sub_agent and active_widget state.
  • Image and page attachments are normalized into structured context so the right workflow owns the turn.

Guardrails and routing

Production AI in commerce needs hard boundaries, not only prompt instructions.

  • Routing prioritizes explicit sub-agent ownership first, then widget-derived specialist mapping.
  • Branding completion, plan gating, and out-of-scope checks prevent unsupported flows from leaking through.
  • Max Insights turns structured store data into natural-language coaching with optional chart generation when the user explicitly asks for it.

Highlights

What shipped

A production AI copilot that routes merchants between a general orchestrator and specialist agents for insights, merchandising, copywriting, and page building.

  • Worked on a unified copilot that connects branding and media tools with specialists for analytics, merchandising, copywriting, and page building.
  • Implemented streaming chat behaviors that keep widgets, approvals, pickers, and longer-running actions coherent in the UI.
  • Built guardrails around branding readiness, plan access, active-agent ownership, and publish-safe workflows.
  • Integrated live analytics context so merchants can ask about orders, traffic, leads, sales, and product performance in natural language.

Implementation details and screenshots are intentionally limited because this is active product work.