Cloud billing is opaque by design. Each provider gives you an invoice optimized for their revenue recognition, not your capacity planning.

Run workloads across AWS, GCP, and Azure and you get three different billing models, three different SKU taxonomies, three different ways of saying "compute." Finance reconciles spreadsheets. Engineering guesses. Workload placement happens on hope.

I needed one view: footprint, cost, trend, forecast. Across vendors. Without paying $50k/year for a FinOps platform.

THE STACK

FOCUS (FinOps Open Cost and Usage Specification) normalizes cloud billing across providers. Ingest into TimescaleDB hypertable — handles millions of line items, months of data in gigabytes. Hasura for instant GraphQL endpoints. deck.gl for WebGL-accelerated visualization.

$50/month infrastructure. Few hours setup. Non-intrusive — just an API token for the FOCUS endpoint.

WHAT MADE IT USEFUL

Cost data alone doesn't tell you where to run things. So I added business context.

Bureau of Economic Analysis + Census integration: Direct API pulls for population and GDP by metro region. Auto-refreshed.

Latency Rings: Pick a region, set a threshold — say, 50ms. The viz shows how many people this workload can serve under that latency, and their GDP profile.

Now the question isn't "which region is cheapest?" It's "how many people can we serve under X ms, what's their economic value, and what does it cost?"

us-east-1 isn't the default because it's best. It's the default because nobody did the math. This does the math.

There's a "Reload data" button to flush from FOCUS APIs. Goal is 15-minute granularity — catch misuse on the fly, not in next month's invoice.

WHERE FOCUS FALLS SHORT

Commitment attribution — RIs and savings plans don't map cleanly to workloads. Tag hygiene — garbage in, garbage out. Network and egress — still a mess across providers.

---

The question that used to take a meeting: "How much are we spending in us-west-2 and how many users are we serving top-notch service?"

Drill down on the region, click a few buttons. Five seconds. Answer.

Demo: https://ure.us/atlas

I spent a few hours on this to solve a specific need. Turned out cooler than expected.

If there's interest, I'll open-source it — ingestion scripts, TimescaleDB schema, Hasura config, BEA/Census integration, latency ring logic, deck.gl dashboard.

Let me know.