The Infinite Library interrogates the limits of algorithmic recommendation applied to literature. This project, built with Next.js and a vector database architecture, reproduces and extends literature-map.com concept by substituting user behavioral data with a vector embeddings system meant to capture similarities between authors. The technical principle is simple: language models transform author information into multidimensional numerical representations, enabling the calculation of mathematical proximities between writers*. These distances in vector space are then translated into recommendations, creating a cartography of literary affinities according to the machine.

The result produces sometimes surprising connections, and we must be clear about what these connections reveal: these links are artificial. They proceed from no literary analysis in the traditional sense, rely on no knowledge of historical contexts, acknowledged influences, or aesthetic debates that actually traverse the works. The algorithm operates on linguistic patterns, lexical recurrences, syntactic structures, producing associations that say more about the biases of training corpora than about genuine literary kinships. Yet these mechanical approximations present a singular documentary interest. They offer a window into how machines process language, the implicit categories they construct, their particular way of reading without understanding.

This project therefore explores less literature itself than the technical mediations that now interpose between readers and works, questioning what we accept to delegate to machines in our relationship to culture.

*Given my humble literary background, I have to admit the calculations mentionned were also generated with the help of an LLM.