Posts

An interactive overview of AI agents and LLMs, with filters by category, pricing, and use cases. The page below is the embedded version of assets/data/index.html.

Most bad coding-agent prompts are too short in the wrong places. They say what to add, but not what scientific contract must remain true. For research code, that is risky. A small implementation detail can change a convergence rate, a sign convention, a nondimensional number, or the meaning of a validation result.

Fine-tuning Zephyr-7B for Finite Volumes: Methodology and Results