I ran 6 experiments with real API calls to benchmark prompt caching. Without optimization, most developers get 0% cache hits without realizing it. CacheGuardian fixes this with one line of code.
Implementing the GIST algorithm from NeurIPS 2025 as a production-grade Python package for selecting high-quality, diverse subsets from large datasets using submodular optimization.
How I built Colander, a production-grade HTTP caching reverse proxy in Rust that replaces LRU with the SIEVE eviction algorithm for lock-free cache hits and near-linear multi-thread scaling.