MMMaurizio Morriinmauriziomorri.hashnode.dev·May 29 · 5 min readCan a Model Read a Cell Like a Sentence?For the last few years, a strange and powerful idea has been working its way through cell biology. The idea is this: if you squint, a cell looks a little like a sentence. A sentence is made of words a00
MMMaurizio Morriinmauriziomorri.hashnode.dev·Mar 13 · 4 min readWhy Planning Is Becoming More Important Than AutocompleteThe most interesting programming story of the last two weeks is not just that models got better at writing code. It is that the center of gravity is moving from raw code generation toward planning, re00
MMMaurizio Morriinmauriziomorri.hashnode.dev·Mar 10 · 3 min readFour Tricks That Make Long Context Inference Actually Work in ProductionMost performance talk about large language models still fixates on raw compute, but long context serving is usually a memory problem first. During decoding, the model must reuse the key value cache fo00
MMMaurizio Morriinmauriziomorri.hashnode.dev·Mar 4 · 3 min readKV Cache Is the Real Bottleneck in Long Context InferenceIf you have been benchmarking LLM inference and wondering why latency and batch size collapse as context grows, the answer is usually not FLOPs. It is memory, and specifically the key value cache. Dur00
MMMaurizio Morriinmauriziomorri.hashnode.dev·Feb 20 · 3 min readSecure CI CD in 2026: Provenance, Signing, and the Pipeline as a Threat Model If you run CI CD the same way you did a few years ago, you are probably shipping faster than ever and also trusting more invisible machinery than ever. The pipeline is now part of your threat model. T00