NYIT builds an LLM assistant for its applied analysis course in MATLAB

Researchers in the Department of Electrical and Computer Engineering Technology at the New York Institute of Technology (NYIT) presented VectorizationLLM on arXiv in July 2026. It is a narrowly specialized language assistant built on Google's open-weight models and designed for one specific course.

Which course the assistant was built for

VectorizationLLM was developed exclusively for CTEC 247, "Applied Computational Analysis II," offered by the ECE Technology department on NYIT's Old Westbury campus in New York. The course covers several topics that engineering students traditionally find difficult:

• Smart vectorization in MATLAB
• Analysis of time and wave vectors
• Piecewise-defined functions
• Fourier analysis
• Ordinary differential equations

These are the topics where students most often get stuck and look for help outside the classroom. The assistant fills that gap by being available around the clock, unlike an instructor in the department.

How the model works, and what it does not do

VectorizationLLM's architecture has two components. The first is RAG (Retrieval Augmented Generation): a knowledge base built from CTEC 247 course materials, including lecture notes, MATLAB code examples and illustrations. When a student asks a question, the model first searches this corpus for relevant passages, then generates an answer based on them. This sharply reduces the risk of "hallucinations" compared with general-purpose chatbots.

The second component is a system prompt that sets a pedagogical role: the assistant explains and gives hints, but deliberately does not provide ready-made solutions to problems. The principle is a "Socratic dialogue": the student has to reach the answer on their own. Responses can contain text explanations, code blocks and images at the same time, which lets the assistant explain mathematical concepts in the most visual format.

Why a "course-specific AI" is a promising direction

VectorizationLLM fits into a growing trend toward narrowly specialized educational LLMs tied to a particular subject or course. Unlike raw ChatGPT, such an assistant stays within the curriculum, relies only on the instructor's verified materials and reduces the department's workload of routine student questions.

The key challenge in replicating such solutions is filling the RAG corpus: building a quality knowledge base for each course takes time and academic expertise. That, rather than the LLM itself, is where most of the work lies.

What this means

VectorizationLLM shows how Google's open-weight models can be used in education without costly fine-tuning. Combining RAG with a system prompt makes it possible to adapt a base LLM quickly to any course while keeping control over the content. Wide adoption of such tools at universities will depend on how easily instructors can build these knowledge bases themselves.