Key results
The company
iFood
ifood.com.brOnline food delivery and grocery marketplace platform in Brazil.
Result highlights
- ~70k developer hours saved per month
- 33% reduction in global code review time
- ~7 hours saved per auto-approved review
- Small MR review time cut from 12 hours to 5 hours
- Auto-approval completed in <1 minute
- 7k code reviews auto-approved monthly
The story
A leading Latin American delivery platform processes over 160 million monthly orders across food, grocery, and fintech services with a team of 1,400 engineers.
Code reviews created a significant bottleneck, with developers waiting an average of 65 hours for approval even on small updates under 30 lines. This delay missed internal targets by nearly two days and prevented new features from reaching customers quickly.
The engineering team integrated Gemini 2.5 Flash into their internal developer platform to function as a binary risk analyzer rather than a standard code assistant. This system evaluates metadata and documentation to instantly approve low-risk updates under 30 lines, bypassing human review entirely for safe changes. The architecture uses strict output schemas and latency controls to ensure security compliance while processing thousands of requests monthly.
Quotes
“A escolha pelo Google Cloud, especificamente pelo Vertex AI neste projeto, foi crucial por oferecer um modelo de IA generativa como o Gemini 2.5 Flash, que era, de longe, o mais eficiente em custo-benefício.”