Levels.fyi

The founders launched Levels.fyi without a traditional database or API server. Submissions came through Google Forms; data lived in Google Sheets; Lambda plus API Gateway handled writes; reads were baked into JSON on S3 and served through a CDN.

Their engineering blog is explicit: this let them focus on product-market fit instead of provisioning RDS instances they did not yet need.

The architecture scaled to millions of users and worked for roughly 24 months, according to their post. Every salary record could ship to the browser in a single JSON file; charts and stats ran client-side.

Pain arrived as files grew to multiple megabytes, Lambda timed out on big transforms, and they hit Google Sheets API rate limits — plus the 10-million-cell cap, which they temporarily worked around by sharding into multiple sheets.

They migrated reads and writes to Postgres and real APIs while duplicating writes to Sheets during transition — no big-bang rewrite. A January 2024 update notes Sheets still used throughout operations alongside Postgres.

Their philosophy did not flip to "scale everything." One of their most trafficked services still ran on a single Node.js instance serving tens of thousands of requests per hour when they published the story.