wtf is rag-decision-diagnoser?

This project is a single page web app that helps someone decide which approach to use when building an AI system that needs outside information: retrieval augmented generation (RAG), long context, fine tuning, or a mix of these called hybrid. It runs entirely in the browser, with no backend server, no calls to any AI model or API, no environment variables, and no database. Every answer it gives comes from a fixed decision tree written in code, so the same inputs always produce the same verdict. The tool walks a user through a set of questions and then shows a recommendation along with the exact path through the decision tree that led to it, so the reasoning is visible rather than hidden. It also scores how confident it is in that answer, since different signals in the questionnaire can either agree with each other or point in different directions. The decision tree follows a set order of preferences. Long context is favored for a small, single document that needs deep analysis. A hybrid approach is chosen when both the system's knowledge and its behavior need fixing. Retrieval is used when the answer must cite sources, and it escalates to hybrid if the underlying problem is also behavioral. Fine tuning is reserved for behavior problems that are stable over time. The tool also handles cases where retrieval can be skipped entirely, such as a tiny and stable set of facts that fits in a single prompt. A cost comparison feature is included, but the project is upfront that its cost numbers are demonstration assumptions from a product brief, not real vendor pricing or guaranteed benchmark results. The comments in the code point to several published research papers that back up the general shape of these tradeoffs, covering topics like retrieval versus long context, how language models handle long inputs, fine tuning versus retrieval, and faithfulness in RAG citations. Five ready made example scenarios are built in, such as a customer support bot, a brand voice email writer, a legal contract analyzer, a medical assistant needing current research, and an internal company question and answer tool, each resolving to one of the four architecture choices. It can be run locally with npm or deployed with no extra configuration on Vercel.