Using Local AI for Literary Analysis

Justification and Rough Design

Below I've written a paper which is the justification for the Literary Analyst project and the start of my design for a Local AI based Literary Analyst Agent. I have been using Google's Notebook LM to analyze Children of Sophista's "Stefan's Owl from Oblivion" book series. This book series contains eight volumes of dense Post Modern Literary Fiction. It contains very detailed characterizations and numerous philosophical musings concealing in the narrative numerous age-old Philosophical problems disguised as problems the child characters in the series must solve in their daily, child deity lives.

One of the problems with such a large series is keeping plot, theme, and philosophical coherence through all the books. Besides telling me what literary tools I have unknowingly used, the Notebook LM AI can answer questions such as, "Has character A ever told B about C." This is extremely useful in editing and enhancing the series. Such questions help prevent what I call "phase errors," where a character exhibits knowledge that it could have only obtained later in the series. It can also answer questions like, "Does the character reflect a personality that could have evolved from what it was previously?" AI is an indispensable writer's tool.

In the paper below, I give the justification for moving from the cloud based AI Notebook LM to a local AI for most, if not eventually all, literary analysis tasks. For those that want to know, the machine hosting the local AI is a pretty powerful one. It is not a simple laptop or consumer PC AI implementation, but a deep AI implementation requiring powerful hardware support. As a commercial would say, "do not attempt this at home." The machine not only runs Linux AI / AI development tools in a WSL2 environment but the entire Adobe publishing tool chain on the Windows 11 side. The illustration and book layout tasks put a pretty heavy demand on the graphics card. 

Machine Specifications: 

Dell 7960 with 128 GB DDR5 memory,  XEON W5-3435X (16 cores, 32 threads, 4.7Ghz Turbo)
4GB M.2 SSD + M.2 4GB RAID 1 SSD array + 4TB SATA Enterprise HD
Nvidia RTX 4000 Ada 20GB DDR6 used for driving displays and "mini AI" used in RAG (see paper)
Nvidia RTX 6000 Blackwell 96GB DDR7 dedicated for large AI (see paper)
​

Literary Agent Rough Design 6/4/26