Longitudinal Benchmark for learned Indexes

As part of this project, I built a longitudinal benchmark for learned index structures to systematically evaluate their performance over time and under varied workloads.

This involved designing experiments to test fundamental operations such as insertions, deletions, and queries, revealing substantial weaknesses in the current implementations.

A key part of the project was writing concurrent C++ code to simulate real-world usage and measure performance under multi-threaded access patterns. This required careful handling of synchronization, race conditions, and efficient memory management to ensure accurate and reproducible results.

The project provided deep experience in systems-level programming, benchmarking, and performance analysis, as well as insights into how machine learning techniques can be integrated into traditional data structures while identifying practical limitations.

Skills and technologies used: C++, concurrent programming, benchmarking, learned index structures, performance analysis, multi-threaded systems, experimental design.