Collaborative Workspace

Build a workspace where users upload documents (PDF, PPTX, XLSX), chat with them, run agents against them, and produce reports. Multiple users can collaborate in the same workspace.

Core requirements

• Upload PDFs, PowerPoints, and Excel files to a workspace
• Extract content from uploads to enable search over documents
• Chat interface where users ask questions against one or more documents
• Ability to trigger an agent run that produces a report from workspace documents
• Multiple users can be members of a workspace and see each other's uploads/chats

What we're evaluating

Use whatever coding agent you'd like (Claude Code, Cursor, etc.) — we care about the judgment behind what you build, not whether you typed every line by hand. Be ready to explain what you built, where you trusted the agent, and where you overrode it. There's no single correct architecture; make reasonable tradeoffs and be ready to defend them, including how the design would change under load (100k users, 5GB files, concurrent uploads, live collaboration, one workspace with a million documents).

Structure

/frontend       # UI: upload flow, chat, loading/error states
/backend        # API, background jobs, schema
/infra
  architecture.md   # describe your infra choices (no IaC needed)

infra/architecture.md should describe, in prose/diagrams: hosting, storage, background job/queue setup, database choice, and how you'd sandbox any code execution (e.g. agent-generated Python) — containers, resource limits, network isolation, cleanup. No Terraform required, just the reasoning.

Not in scope

• Auth/SSO polish (a stubbed login is fine)
• Production-grade CI/CD
• Algorithmic puzzle-solving — this is a product-building exercise