05 Development notes / architecture / releases

Lotargo Blog

Notes about AI systems, RAG architecture, runtime experiments, verification work, and the decisions behind public project pages.

Latest notes

A static archive that can later grow from markdown sources into generated pages.

Back home

The Filter Matryoshka: Notes on Gemini Safety LayersМатрёшка фильтров: заметки о safety-слоях Gemini

A defensive note about input moderation, reasoning traces, final-output filters, and why LLM safety systems should not depend on a single layer.

Defensive-разбор input moderation, reasoning-трасс, финального output-фильтра и того, почему безопасная LLM-система не должна зависеть от одного слоя.

Read note