Build an AI spend dashboard workflow that consolidates our company's AI tool spending across multiple providers into one browser dashboard, with per-user and per-department attribution. The goal is to let a team see, in one place, how much they're spending on AI coding and API tools, broken down by source, user, model, and department, and how usage shifts over time. Triggering and data collection. Create a single entry step, "Refresh spend data," that runs on an hourly cron and can also be hit on demand via an authenticated route. It fans out to a set of fetch steps that each pull spend from one provider and write a normalized daily file ({ date, email, source, model, cost_usd }) to a shared named volume. Build these fetchers: Cursor spend — pull daily usage events and the current-cycle billing split (included vs. on-demand) from the Cursor Admin API. Use Cursor here, or whatever AI coding tool your team uses. Claude/Anthropic API spend — pull token usage from Anthropic's Usage & Cost Admin API, mapping users and API keys to emails for per-user attribution, and estimating cost from a per-model price table. Swap in whichever LLM API provider you use. Cloud LLM spend (Bedrock) — pull per-user model spend from AWS Cost Explorer, filtered to the Bedrock service. Use this, or your cloud provider's cost/billing API. Department directory — pull an employee email → department map from BambooHR (or your HRIS) and maintain it as a sticky, append-only map so departed employees keep their attribution. Make the fetchers incremental (re-fetch only recent days and merge) with a full backfill when the stored file is missing. Store raw provider model strings and normalize them to canonical labels at read time, so label changes never require a re-fetch. Serving and display. Add a "Spend data API" step that reads all the source files, normalizes models, joins the department map, and returns compact day-level cells (email × source × model × cost) plus each source's freshness and the billing split. Add a "Dashboard" step — a client-side React page on an authenticated route that renders immediately and fetches from the data API. It should offer filtering by department, user, source, and model, a time-series trend chart with grouping toggles, per-user breakdowns, richer exploratory views (flow/Sankey, streamgraph, treemap), model-migration/loyalty views, token-usage views, and CSV export. Optionally add a Cursor deep-dive analytics endpoint for per-request token/context stats. Keep all routes space-authenticated. Give it a polished, distinctive, on-brand design. Tools used: Cursor Admin API, Anthropic Usage & Cost Admin API (via an admin read-only workflow-connector proxy), AWS Cost Explorer (Amazon Bedrock spend), BambooHR employee directory, React (client-side dashboard), 3B named volumes.