<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Writing · José Manuel Ramírez Sánchez</title><description>Essays and analyses on speech, health and the safety of generative AI. Each one links to the tutorial or papers behind it.</description><link>https://jmramirez.engineer/</link><language>en</language><item><title>Coding agents, the basics: Claude Code for healthcare professionals</title><link>https://jmramirez.engineer/writing/coding-agents-basics/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/coding-agents-basics/</guid><description>How a coding agent like Claude Code works and how to keep it under control, told through one fictitious HbA1c report: rules, context, subagents, skills, privacy and which model to use.</description><pubDate>Wed, 11 Nov 2026 00:00:00 GMT</pubDate><category>Coding agents</category><category>Claude Code</category><category>Healthcare</category><category>AI agents</category><author>José Manuel Ramírez Sánchez</author></item><item><title>Two second places at ALBAYZIN 2026: recognising and diarizing rural Spanish</title><link>https://jmramirez.engineer/writing/albayzin-2026-coser/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/albayzin-2026-coser/</guid><description>My systems finished second in both the speech recognition and the speaker diarization tracks of the ALBAYZIN 2026 Speech-COSER challenge, on dialectal rural Spanish, trained on nothing but the challenge&apos;s own COSER data.</description><pubDate>Sat, 31 Oct 2026 00:00:00 GMT</pubDate><category>Speech recognition</category><category>Speaker diarization</category><category>Evaluation campaigns</category><category>Whisper</category><category>Parameter-efficient fine-tuning</category><author>José Manuel Ramírez Sánchez</author></item><item><title>Radial LoRA: when the direction matters more than the size</title><link>https://jmramirez.engineer/writing/radial-lora/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/radial-lora/</guid><description>R-LoRA is a variant of LoRA that tempers the size of the adapter&apos;s input with one learnable exponent per layer. On four clinical speech corpora it beats LoRA by 12.4 points of Macro-F1, and it is a drop-in change for any Transformer that LoRA can adapt.</description><pubDate>Fri, 30 Oct 2026 00:00:00 GMT</pubDate><category>Parameter-efficient fine-tuning</category><category>LoRA</category><category>Speech biomarkers</category><category>Clinical speech</category><category>Transformers</category><author>José Manuel Ramírez Sánchez</author></item><item><title>The 10% of the terminal that solves 90% of the job</title><link>https://jmramirez.engineer/writing/terminal-10-90/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/terminal-10-90/</guid><description>The four terminal skills you will use the most: moving around without a GUI, SSH with an alias for you and your coding agent, tmux, and watching the CPU and the GPU.</description><pubDate>Wed, 28 Oct 2026 00:00:00 GMT</pubDate><category>Terminal</category><category>Linux</category><category>SSH</category><category>tmux</category><category>Coding agents</category><author>José Manuel Ramírez Sánchez</author></item><item><title>How LLM guards decide: a tour inside four safety models</title><link>https://jmramirez.engineer/writing/how-llm-guards-decide/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/how-llm-guards-decide/</guid><description>Where the policy lives, which layers read it, what the router does and why the last block can overturn a verdict: a tour through the internals of four LLM guards, measured on the weights they actually run.</description><pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate><category>AI safety</category><category>LLM safeguards</category><category>Interpretability</category><category>Transformers</category><category>Mixture of experts</category><author>José Manuel Ramírez Sánchez</author></item><item><title>Inside four LLM guards: what the weights say about safeguards for agentic systems</title><link>https://jmramirez.engineer/writing/inside-four-llm-guards/</link><guid isPermaLink="true">https://jmramirez.engineer/writing/inside-four-llm-guards/</guid><description>I audited four candidate guards for an agentic assistant layer by layer. They all decide in one token and late in the network, and the two dedicated guards miss prompt injection hidden in documents. A small generalist catches almost everything if it writes its justification before the verdict.</description><pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate><category>AI safety</category><category>LLM safeguards</category><category>Prompt injection</category><category>Interpretability</category><category>Agentic AI</category><author>José Manuel Ramírez Sánchez</author></item></channel></rss>