How Tinder built and open-sourced Bazel-diff to transform their CI/CD at scale

Join us with Maxwell Elliott and Connor Wybranowski from Tinder iOS platform team as they share how their small team of 5 transformed CI/CD bottlenecks affecting 40+ engineers across the organization.

Facing 40-60 minute CI builds on a 1.5 million line codebase with 300+ targets, Tinder's team developed Bazel-diff, an open source tool that uses Merkle tree cryptographic summation to provide 100% accurate target diffing between Git revisions. The staggering results: Tinder saves over 4 years of compute time annually in their P90 case and 17 years in their P100 case.

Combined with Buildkite's dynamic pipelines, preliminary data shows potential 93% time savings in P100 scenarios, a 54% improvement over Bazel-diff alone. This allows them to decompose monolithic build processes into distributed, optimized pipelines that scale horizontally.

In this technical discussion, we’ll cover:

• How Bazel-diff's Merkle tree architecture enables precise target selection at scale
• Why they built it as an open source project and how the community contributed
• The technical synergies and benefits of integrating Bazel-diff with Buildkite's dynamic pipelines
• How they moved from monolithic builds to distributed shards for better resource utilization
• Why this approach matters for AI-driven development workflows
• Practical implementation advice for teams adopting this combination

Discover how Tinder transformed their CI/CD pipeline from frustration into a competitive advantage and how your team can implement similar improvements.