Kmemo 2.1.0, a semantic cache for LLM calls that r...

# feed

[Antonio S]

Kmemo 2.1.0, a semantic cache for LLM calls that refuses to serve the wrong answer. "Convert 100 USD to EUR" and "Convert 250 USD to EUR" embed at 0.99, so eleven lexical guards read the candidates as text instead of trusting the threshold.

Since 1.0 the core went Kotlin Multiplatform: JVM, iOS, macOS, Linux, Windows, JS and Wasm, still with kotlinx-coroutines-core as its only dependency.

2.1 measures those guards against a corpus I did not write, PAWS from Google Research. They do much worse on it: 14% of near misses caught, against roughly 68% on my own blind splits. It ships in the README beside the good numbers.

Apache-2.0. github.com/NaCode-Studios/Kmemo