Nubank uses fine-tuned Devin to split a 6M-line ETL monolith, claiming 8–12x engineering-hour efficiency
Nubank needed to split an 8-year-old ETL monolith of more than 6 million lines into sub-modules: about 100,000 data-class migrations, scoped as an 18-month effort across 1,000+ engineers. Cognition fine-tuned Devin on Nubank's earlier manual migrations, ran many instances in parallel, and engineers reviewed and merged the PRs. The Data, Collections and Risk units finished their migrations in weeks.
The numbers (as reported)
| Codebase | 6M+ lines, ~100,000 data-class implementations |
|---|---|
| Original estimate | 18 months, 1,000+ engineers |
| Engineering-hour efficiency | 8–12x |
| Cost savings on delegated scope | 20x+ |
| Time per sub-task after fine-tuning | ~40 → ~10 minutes |
Facts
- Company / subject
- Nubank
- Tools
- Devin (fine-tuned)
- Codebase
- brownfield
- Work type
- Migration
- Stack
- ETL, Data pipelines
- Published
- 2024-12
- Last checked
- 2026-10-18
Caveats
Vendor-published. The page states 8x, 8–12x and 12x in different places. Efficiency compares typical manual hours with hours spent prompting and reviewing. Cost compares Devin's run cost with engineer hourly cost against an estimated manual baseline. The work was highly repetitive and suited to agents. Needed custom fine-tuning, and early weeks had frequent errors to fix. The publish date comes from Cognition's LinkedIn announcement (Dec 2024), not the page.