Design the search backend for a team workspace: one search box that spans everything — documents, comments, tasks, and file names — returns ranked, permission-correct results scoped to the searcher's workspace, and feels instant as they type. It is the unified index over heterogeneous collaboration content.

Operating context. Content changes constantly (new docs, edits, deletions, task updates) and must become findable within seconds. Every result is subject to permissions: a user must never see a snippet of something they cannot access, even for the moment right after a permission changes. Queries are workspace-scoped and skewed toward a few large, active workspaces. Typeahead suggestions fire on nearly every keystroke, so that path is even more latency-sensitive than full search.

Out of scope. The content stores themselves (docs, chat, tasks own their own storage), authentication, the query-box UI, and semantic/vector retrieval — assume classic keyword search for v1. A change-event stream from each content system is available to you.

What to produce. A high-level architecture covering: the index architecture and sharding across millions of workspaces; the ingestion pipeline that turns content changes into near-real-time index updates; permission-aware retrieval that never leaks a forbidden result even as ACLs change; ranking and relevance across mixed entity types; and the low-latency typeahead path. Sketch the components and the flow of both an index update and a query; we will probe specifics in checkpoints.