Build an internal team people directory as a 3B workflow. Purpose: on growing or distributed teams, people don't really know each other's skills, projects, or interests, so connections that should happen never do. This workflow gives everyone a searchable profile and uses an LLM to suggest teammates they'd click with. Trigger: HTTP routes. A single web page is the whole UI, served at its own route and space-authenticated. Identify each user automatically from the platform's spoof-proof authenticated-email header — no separate login or signup, and never trust a submitted email. High-level steps: A web app page that renders immediately and fetches its data from a few small JSON API endpoints. An endpoint returning the current caller's identity. An endpoint listing all saved profiles. An endpoint that upserts the caller's own profile (photo, role, skills, projects, hobbies, interests, links). A "discover" endpoint that computes shared-interest overlap between people and, on demand, calls an LLM to write natural-language teammate suggestions. Only run the LLM when the roster has actually changed, and always return the overlap-based matches even if the LLM call fails. Endpoints to save and retrieve each user's history of generated suggestions. Storage: persist profiles and suggestion history in a shared database on a named volume, with the write endpoints as the exclusive writers and read endpoints mounting read-only. External services: use an LLM for the discovery suggestions — Anthropic's Claude via a connector, or your team's preferred model provider. Everything else is self-contained. Keep access private to the space unless asked otherwise. Tools used: 3B, SQLite, Anthropic