Lead product designer on the client side • Project duration 9 months
Disclaimer
This project is under NDA, so I'm showing the process and key decisions here. Sensitive information, client name, and proprietary data have been generalized. I'm happy to walk you through the full case, screens included, on a call.
Client overview
June is an AI collaboration platform where teams work with AI experts, create documents together and run multi-step processes.
Scope
Onboarding, dashboard and experts hub · files & folders sections · multi-expert chat · processes and templates · shared workflows (mentions, approvals, version history) · artifact editors (documents, slides, maps, tables) · admin and role-based access · design system
Problem
The product started as a chat with AI. The client wanted a system where several people could work on the same documents and processes, with AI experts helping along the way.
How I worked
Brief breakdown → questions to the client → desk research → quick sketches for early feedback → user stories → review cycles with the product owner and developer → component cleanup, dev-ready screens and clickable prototypes.
Key decisions
From chat to a collaborative system
Challenge: A chat works for one person asking questions, but not for a team building documents together. Several people editing the same artifact means conflicts and lost changes.
Decision: Based on a UX audit, I redesigned the product around processes made of steps, each producing an artifact. For parallel work, I applied Git logic to documents: branches, commits, merging, version history, and request-to-edit with locking.
Orchestrator: turning any planning form into an AI template
The Orchestrator turns a planning form (PDF, DOCX, slides) into a reusable AI template. It splits the document into cards, and each card holds a field, an instruction for the AI, and the sources and tools to use. The original interface was a bare form with a drop zone.
Challenge (1): Users couldn't see which part of the document a card came from, or what was already done.
Decision: I split the flow into setup → run → review. In setup, the form sits on the left with a pin on each section, and clicking a pin highlights that area. In review, accepted sections get a checkmark on their pin, and the right panel turns into a chat, consistent with patterns users already know from other parts of the product.
Challenge (2): Templates built from scratch have no document, so the split view left half the screen empty, and the client flagged it.
Decision: I explored a generated form preview, a template summary and a single column. In the end, the cards moved to the left and the right panel became the space for adding new ones. In review without a document, the chat takes the full width: with nothing to anchor to, a split view makes no sense.
Challenge (3): In review stage, the user has to quickly find what needs their attention.
Decision: The AI surfaces missing and conflicting sections first. A missing field can be typed in or handed back to the AI with a prompt. A conflict shows the value from each source, and the user picks one or ignores it. Each section is confirmed separately with Accept or Retry.
Impact
The platform moved from a single AI chat to a full collaborative system. All core flows were built and are now live in production.