How We Use AlphaEvolve to Make Complex IDE Algorithms Faster | The JetBrains AI Blog

AlphaEvolve, a Google DeepMind algorithm-discovery system, was utilized by JetBrains to enhance the indexing speed of IntelliJ-based IDEs, a critical feature impacting navigation, search, and other code insights. By leveraging Gemini to generate and refine algorithm improvements, AlphaEvolve aimed to identify optimization candidates within the already optimized B-tree structure of the IDEs' indexing infrastructure. The experiment focused on validating these candidates through synthetic benchmarks and full IDE integration tests, resulting in a 15-20% improvement in synthetic performance scores across sessions with more than 50 iterations. However, only one out of five generated candidates demonstrated a statistically significant improvement in real-world IDE scenarios, reducing the indexing time from a baseline of 17.4 ± 0.5 seconds to 16.6 ± 0.2 seconds. This approach highlighted the importance of integrating autonomous search with practical validation to ensure user-visible performance enhancements, underscoring AlphaEvolve's role in generating and ranking low-level optimization ideas in a space where manual exploration is challenging. The next steps involve product validation to assess improvements in the Mega Index metric, which tracks indexing performance and user satisfaction.