https://mpklu.github.io/tags/structured-output/Recent content in Structured-Output on AI Coding BlogHugoen-usMon, 06 Apr 2026 12:00:00 -0400https://mpklu.github.io/posts/gemma-4-benchmark/Mon, 06 Apr 2026 12:00:00 -0400https://mpklu.github.io/posts/gemma-4-benchmark/<h1 id="gemma-4-structured-task-performance-field-report-from-a-local-first-app">Gemma 4 Structured-Task Performance: Field Report from a Local-First App</h1> <p><em>Benchmark data and prompt-format findings from deploying Gemma 4 E4B in a real application. Intended for LLM teams (Gemma, Ollama) and developers building structured-output pipelines on local models.</em></p> <hr> <h2 id="context">Context</h2> <p>We build <a href="https://github.com/mpklu/gary">Gary</a>, a privacy-first personal assistant CLI for macOS. It runs entirely locally — an encrypted database, a daemon process, and a local LLM via ollama. The LLM handles three structured tasks:</p>