ahic shows a simple RAG build on Gemini API and Python

The guide's design deliberately skips the machinery of a full RAG pipeline. Instead of embedding documents into vectors and retrieving relevant chunks, it simply concatenates every .txt file in a 'documents' folder into one large context and hands it to the model with strict instructions. This trades scalability for setup speed — no indexing, no vector database — which is why the author positions it as a 'simple version' suited to shorter collections.

The two examples the author ships with illustrate the intended use: an internal guideline covering leave requests, expense deadlines, remote-work limits and hiring questions, and a project memo listing the development phase end date, key customers, and the next regular meeting. Questions like how many days ahead a leave request is needed, or when Project A's next meeting is, are answered straight from those files, while a question about salaries is refused because it isn't covered.

The author also flags where this breaks down. The Gemini API caps how many tokens can be processed at once, so very long or numerous documents can trigger errors once everything is concatenated. The suggested path forward is chunking documents, embedding them, and retrieving only the most relevant chunks — the standard full RAG approach this simple version intentionally avoids.