How dogfooding helps us build a better Duolingo
Article Summary
Duolingo catches critical bugs before users ever see them. Their secret? 70% of employees actively use internal builds daily.
Blanca Zhang from Duolingo shares how the company turned dogfooding into a systematic quality assurance process. They've built custom tooling to extract maximum value from internal testing across their entire organization.
Key Takeaways
- Release Dashboard surfaces ANR, crash, and OOM metrics within hours of new builds
- Shake-to-Report captures bugs with Fullstory recordings and automatic metadata
- AI tool Jeeves aggregates feedback spikes from dogfooding and external sources
- Caught Music course crash bug weeks before launch using enhanced logging
- No Monday rollouts happen until QA clears all blocking dogfooding bugs
By combining company-wide dogfooding culture with purpose-built tooling, Duolingo identifies and fixes critical issues before they reach production.
About This Article
Duolingo needed to catch performance regressions and crashes across different app versions and device types before they reached users. When the Music course launched, crashes hit nearly 2,000 events and required fast diagnosis.
Blanca Zhang's team built the Release Dashboard to show telemetry metrics within hours of new builds entering dogfooding. They also improved logging in bug reports so engineers could see detailed course interaction data and trace problems back to their source.
Engineers looked at detailed logs from Shake-to-Report bug submissions and found an edge case in how songs paused. They fixed it weeks before the official iOS launch, which kept the problem from affecting a lot of users.
