Research

Deep-dive technical memos, architecture audits, and quantitative insights.

2026-04-05

The Case Against O(N²) LLM Wiki Updates

Why academic elegance fails in production: Dissecting the token-burn and cascading failure risks of autonomous AI knowledge base maintenance, and proposing a dual-core async architecture.

ArchitectureLLMs
2026-04-06

Surviving the Hardware Lottery: MXFP8 vs M1

A technical analysis of why Edge AI inference fails on older ARM hardware. Understanding the gap between legacy INT8 uniform scaling and modern Microscaling FP8 for LLM reasoning capabilities.

HardwareQuantization