VectorizationLLM: Google-based LLM for MATLAB vectorization

Ryan Duke submitted VectorizationLLM to arXiv on 8 Jul 2026 as arXiv:2607.07846, a 44-page paper with six figures describing a specialized Large Language Model built on Google open-weight LLMs. The model is aimed at students in CTEC 247: Applied Computational Analysis II at the Department of Electrical & Computer Engineering Technology, New York Institute of Technology Old Westbury.

What is VectorizationLLM?

VectorizationLLM is a course-focused LLM designed to assist students learning MATLAB topics, specifically smart vectorization, time/wave vector analysis, piecewise functions, Fourier analysis, and differential equations. The paper presents the model as an instructive assistant that provides detailed explanations and examples drawn from in-class notes while avoiding direct answers to questions.

The abstract lists the target topics and clarifies the pedagogical intent: the model supplies examples in code, text, and images and is designed to be instructive rather than answer giving. The arXiv entry identifies the course application as CTEC 247 at NYIT Old Westbury and notes the manuscript length and figure count: 44 pages and six figures.

How does VectorizationLLM work?

VectorizationLLM pairs a Google open-weight base LLM with a retrieval augmented generation knowledge base and a system prompt architecture to produce course-specific explanations and examples. The paper describes a RAG knowledge base plus system-prompt scaffolding as the core design choices, and it explicitly states that examples in responses include code, text, and images.

The source material frames the pipeline around three elements present in the submission: the open-weight Google model as the backbone, a RAG knowledge base to fetch course material, and a system prompt architecture to shape instructor-style responses. The author emphasizes educational constraints: the assistant presents detailed conceptual explanations and examples taken from in-class notes while not providing direct solutions to students’ questions, quoting the abstract phrase "without providing direct answers to questions." The arXiv record also links to auxiliary sections for code, data, and media associated with the article.

Why it matters

A course-specific LLM that mixes a retriever with a system prompt changes the balance between automated help and pedagogy: it can surface targeted examples and worked code while intentionally withholding final answers, which may preserve learning incentives. For instructors running MATLAB-based computational analysis courses, that pairing offers a reproducible template for embedding model-driven assistance into an existing syllabus.

The paper ties the model to a concrete classroom (CTEC 247) and supplies tangible artifacts (44 pages, six figures) rather than only high-level claims, which makes evaluation and adoption by peers more straightforward.

What to watch

Monitor the DataCite registration noted on the arXiv page: the arXiv-issued DOI is listed as via DataCite with registration pending. Also check the paper’s "Code, Data and Media" section and the linked toggles for concrete demos, notebooks, or source that would let instructors validate the RAG+system-prompt approach in practice.

Citation and submission details: VectorizationLLM, Ryan Duke, arXiv:2607.07846, submitted 8 Jul 2026; paper length 44 pages, six figures; course target CTEC 247 at New York Institute of Technology Old Westbury.