42

This repository is a small collection of solutions to recurring problems that come up when building systems powered by large language models. The author's premise is that many of these problems are structural, meaning they come from how LLMs work at a fundamental level, and no amount of clever prompting or new techniques will fix them if the underlying approach is wrong.

The README uses a pointed analogy: engineers keep reaching for new tools, but they are essentially trying to get a fish to climb a tree by painting it a different color. The problem is not the technique; it is the starting point. The solutions offered here are framed as "foundations" because they aim to address the root causes rather than symptoms.

Each solution in the repository is described as working with deterministic reliability, which means the outcomes should be consistent and predictable rather than varying from run to run. This is a meaningful distinction for anyone who has struggled with LLM-based systems that behave differently on repeated inputs.

The README suggests two ways to use this resource. You can read through the solutions yourself and apply the ideas to your own system by hand. Alternatively, you can point a coding agent at the repository and have it go through the material and apply the relevant solutions automatically. The second path is aimed at developers who already work with AI coding assistants.

The repository is sparse in terms of visible code or detailed documentation beyond the README, so the depth of each solution is something you would discover by exploring the files directly. It is positioned more as a reference or pattern library than a ready-to-install package.